Fluorescence Spectroscopy In Vivo€¦ · FLUORESCENCE SPECTROSCOPY IN VIVO 1 Fluorescence Spectroscopy In Vivo Nirmala Ramanujam University of Wisconsin, Madison, USA 1 Fluorescence

Fluorescence Spectroscopy In Vivo

Nirmala Ramanujam

inEncyclopedia of Analytical Chemistry

R.A. Meyers (Ed.)pp. 20–56

John Wiley & Sons Ltd, Chichester, 2000

FLUORESCENCE SPECTROSCOPY IN VIVO 1

Fluorescence SpectroscopyIn Vivo

Nirmala RamanujamUniversity of Wisconsin, Madison, USA

1 Fluorescence Spectroscopy 11.1 Introduction 11.2 Principles and Definitions 2

2 Fluorophores 42.1 Endogenous Fluorophores 42.2 Exogenous Fluorophores 6

3 Fluorescence Spectroscopy of TurbidMedia 73.1 Fluorescence Spectroscopy of an

Optically Dilute, HomogeneousMedium 7

3.2 Fluorescence Spectroscopy of TurbidMedia such as Tissue 7

3.3 Deconvolution of Absorption andScattering from Tissue FluorescenceEmission Spectra 9

3.4 Turbid Tissue-simulating Phantomsfor Fluorescence Spectroscopy ofTissue 11

4 Instrumentation 114.1 Light Sources 114.2 Illumination and Collection

of Light 124.3 Monochromators and Spectrographs 134.4 Detectors 144.5 Signal-to-noise Ratio Analysis of an

Instrument Used for FluorescenceSpectroscopy of Tissue 15

4.6 Calculation of Tissue FluorescenceEfficiency 15

5 Clinical Applications 165.1 Neoplasia 175.2 Atherosclerosis 19

6 Clinical Instruments 196.1 Single-pixel, Three-excitation-

wavelength Fluorimeter 196.2 Single-pixel, Excitation–Emission

Matrix System 216.3 Fluorescence Imaging Systems 23

7 Methods of Analysis 237.1 Statistically Based Model 247.2 Physically Based Models 25

7.3 Effect of Excitation and EmissionGeometry on Fluorescence EmissionSpectra of Turbid Media 29

8 Future Perspectives 30

Acknowledgments 30

Abbreviations and Acronyms 31

Related Articles 31

References 31

Diagnostic techniques based on optical spectroscopy havethe potential to link the biochemical and morphologicalproperties of tissues to individual patient care. In particu-lar, these techniques are fast, noninvasive and quantitative.Furthermore, they can be used to elucidate key tissuefeatures, such as the cellular metabolic rate, vascularity,intravascular oxygenation and alterations in tissue mor-phology. These tissue features can be interpreted to shedlight on a variety of clinical problems, such as precancer-ous and cancerous growth and atherosclerosis. The goalof this report is to review the development and applicationof optical spectroscopy in the ultraviolet (UV) and visi-ble (VIS) spectral regions, as a diagnostic tool in clinicalapplications. A particular emphasis is placed on steady-state, UV/VIS fluorescence spectroscopy for the detectionof precancers and cancers, in vivo.

1 FLUORESCENCE SPECTROSCOPY

1.1 Introduction

Diagnostic techniques based on optical spectroscopy havethe potential to link the biochemical and morphologi-cal properties of tissues to individual patient care. Inparticular, these techniques are fast, noninvasive andquantitative. Furthermore, they can be used to eluci-date key tissue features, such as the cellular metabolicrate, vascularity, intravascular oxygenation and alter-ations in tissue morphology. These tissue features canbe interpreted to shed light on a variety of clinical prob-lems, such as precancerous and cancerous growth andatherosclerosis..1/ If applied successfully, optical spec-troscopy has the potential to represent an important stepforward toward advances in diagnostic and therapeuticmedical applications.

Spectroscopy is the study of the interaction of elec-tromagnetic radiation with matter. There are threeaspects to a spectroscopic measurement: irradiation ofa sample with electromagnetic radiation; measurementof the absorption, spontaneous emission (fluorescence,phosphorescence) and/or scattering (Rayleigh elastic

Encyclopedia of Analytical ChemistryR.A. Meyers (Ed.) Copyright John Wiley & Sons Ltd

2 BIOMEDICAL SPECTROSCOPY

scattering, Raman inelastic scattering) from the sample;and analysis and interpretation of these measurements.Detailed study of absorption, spontaneous emissionand scattering provides information that can be clas-sified broadly as analytical, structural, dynamic, andenergetic..2/

Optical spectroscopy deals with interactions of elec-tromagnetic radiation with matter that occur at the UV,VIS, near-infrared (NIR) and infrared (IR) wavelengths.In the UV/VIS spectral regions (<700 nm), light can pen-etrate only superficial tissue volumes (a few hundredmicrons to a millimeter in depth) due to the fact that thisbiological medium is highly absorbing..3/ However, inthe NIR spectral region (700–900 nm) tissue is generallyless absorbing and, furthermore, the number of elasticscattering events of light in tissue is approximately twoorders of magnitude greater than the number of absorp-tion events..4/ This enables the light to propagate throughtissue volumes that are up to several centimeters in depth.

In the UV/VIS spectral regions, absorption andfluorescence spectroscopy have been explored extensivelyas diagnostic tools for precancer and cancer detectionin the surface epithelia of various organ sites (colon,cervix, bronchus, lung, bladder, brain, esophagus, headand neck, skin, bile duct, breast and stomach tissues).5 – 12/

and for the characterization of atherosclerosis in thesurface of artery vessel walls..5,6,11/ Recently, fluorescencespectroscopy was also used to detect Alzheimer’s diseaseof the brain in vitro..13/ Elastic scattering spectroscopyin this spectral region also has been evaluated forthe detection of precancers and cancers in the surfaceepithelia of several organ sites, but to a much lesserextent..14/

NIR absorption,.4/ fluorescence and phosphores-cence.15/ spectroscopies have been used to interrogatelarger tissue volumes. In particular, NIR absorption hasbeen used to detect brain bleeds,.16/ brain oxygenation,.17/

functional activity of the brain,.18,19/ bioenergetics ofskeletal muscle (e.g. in the case of genetic disease ofmitochondrial function).20/ and breast tumors..21/ NIRcorrelation spectroscopy, which monitors fluctuations inthe elastic scatter intensity, is emerging as a potential diag-nostic tool to detect blood flow deep within thick tissue..22/

To date, most clinical applications have concentratedon absorption, fluorescence and elastic scattering spec-troscopies, because these measurements can be obtainedwith a good signal-to-noise ratio in reasonably short inte-gration times. However, with advances in illumination anddetection technologies, NIR Raman scattering, which isa relatively weak phenomenon, is emerging as a poten-tial diagnostic tool for precancer and cancer detectionin the surface epithelia of various organ sites,.23/ for thecharacterization of atherosclerosis on the surface of arteryvessel walls.24/ and for glucose monitoring..25/

The goal of this article is to review the developmentand application of UV/VIS optical spectroscopy as a diag-nostic tool in clinical applications. Particular emphasis isplaced on steady-state UV/VIS fluorescence spectroscopyfor the detection of precancers and cancers in vivo. In thissection the principles and definitions that are related tothe phenomenon of fluorescence are summarized. In thenext section the endogenous fluorescence properties ofmolecules present in cells and tissues and the fluores-cence properties of exogenous molecules, which can beused as contrast agents, are presented. This is followed insection 3 by a discussion of the effect of absorption andscattering on fluorescence spectroscopy of turbid mediasuch as tissue, and methods for deconvolving this effect.Next, the instrumentation requirements for fluorescencespectroscopy of tissues are reviewed in section 4. In thelast several sections, a summary of the clinical applica-tions of fluorescence spectroscopy, particularly precancerand cancer detection (section 5), a review of represen-tative instruments (section 6) and methods of analysisused (section 7) are presented. With respect to meth-ods of analysis, statistical models, which are used solelyfor the purpose of discriminating diseased from nondis-eased tissues based on the spectral information, as wellas physical models, which have the potential to eluci-date the biochemical/morphological basis for the spectraldifferences observed, are presented. In the final partof section 7, the effect of the illumination and collec-tion geometry on fluorescence spectroscopy of tissuesis discussed and approaches to resolve it are presented.This article concludes with a brief discussion of futureperspectives in section 8.

1.2 Principles and Definitions

1.2.1 Probing Energy Levels with ElectromagneticRadiation

Optical spectroscopy probes the energy levels of amolecule..2/ The energy level of a molecule is defined asits characteristic state, which is related to the molecularstructure of the molecule and to the energetics anddynamics of any chemical processes that the moleculemay undergo. The ground state of a molecule is definedas the state of lowest energy. States of higher energyare called excited states. A molecule possesses severaldistinct reservoirs of energy levels, including electronic,vibrational, rotational, translational and those associatedwith nuclear and electron spin..2/ In the optical regime,the energy levels of interest are those that are associatedwith vibrational and electronic transitions. The separationbetween vibrational energy levels is determined by themass of the atoms and the flexibility of the chemicalbonds joining them. The separation between electronic


energy levels, which is greater, occurs when electrons aredisplaced from one region of a molecule to another.

At any finite temperature, the molecules will bedistributed among the energy levels available to thembecause of thermal agitation..2/ The exact distribution willdepend on the temperature (T) and on the separationbetween the energy levels (E) in the energy ladder.At a given temperature, the number of molecules inan upper level (nupper) relative to that in a lower level(nlower) is given by the Boltzmann distribution, as shownin Equation (1):

nupper

nlowerD exp

(�EkT

).1/

where k is the Boltzmann constant (1.38ð 10�23 J K�1).When electromagnetic radiation is applied to a molecule,it is just as likely to cause transitions from a higherto a lower energy level as it is to cause transitionsfrom a lower to a higher energy level. Consequently,net absorption or transition to a higher energy level canoccur only if the difference between the populations of theenergy levels concerned is significant, with the lower onebeing significantly higher. Calculation of the populationof energy levels at room temperature.2/ has shown that forvibrational energy level spacings the ratio nupper/nlower is¾10�3 and for electronic energy level spacings it is 10�21.

1.2.2 The Fluorescence Phenomenon

Figure 1 displays an energy level diagram with ground(S0) and excited (S1) electronic states as well as vibra-tional energy levels within each electronic state of amolecule..26/ When a molecule is illuminated at an excita-tion wavelength that lies within the absorption spectrumof that molecule, it will absorb the energy and be acti-vated from its ground state (S0) to an excited singlet state(S1), with the electron in the same spin as its groundstate. The molecule can then relax back from the excitedstate to the ground state by generating energy eithernonradiatively or radiatively, depending on the local envi-ronment. In a nonradiative transition, relaxation occurs

Absorptionexcitation wavelength

Fluorescenceemission wavelength

S0

S1

Nonradiative Radiative

Figure 1 Energy level diagram illustrating the phenomena ofabsorption and fluorescence of a molecule.

by thermal generation (dashed arrows). In a radiativetransition, relaxation occurs via fluorescence at specificemission wavelengths (solid arrow). Fluorescence genera-tion occurs in three steps: thermal equilibrium is achievedrapidly as the electron makes a nonradiative transitionto the lowest vibrational level of the first excited state;the electron then makes a radiative transition to a vibra-tional level of the ground state; and finally the electronmakes a nonradiative transition to the lowest vibrationallevel of the ground state. When there is inter-systemcrossing, in which the spin of the electron is flipped inthe excited state, the time for radiative transition fromthe excited state to ground state is longer because thetransition must occur with a spin change. This excitedstate is termed the triplet state (not shown). Radia-tive transition from the excited triplet state is termedphosphorescence. Tissue absorption, fluorescence andphosphorescence monitor changes in electronic energylevels, to provide biochemical information from biologi-cal molecules.

The phenomenon of fluorescence displays several gen-eral characteristics for a particular biological molecule..26/

First, due to the losses in energy between absorption andemission that occur as a result of nonradiative transitions,fluorescence occurs at emission wavelengths that arealways red-shifted relative to the excitation wavelength.Second, the emission wavelengths are independent of theexcitation wavelength. Third, the fluorescence spectrumof a biological molecule is generally a mirror image of itsabsorption spectrum.

The fluorescence of a biological molecule is character-ized by its quantum yield and its lifetime..26/ The quantumyield is simply the ratio of the number of photons emittedto the number absorbed. The lifetime is defined as theaverage time the biological molecule spends in the excitedstate prior to return to the ground state. The fluorescencequantum yield and lifetime are modified by a number offactors that can increase or decrease the energy losses. Forexample, a molecule may be nonfluorescent as a result ofa large rate of nonradiative decay (thermal generation) ora slow rate of radiative decay (fluorescence generation).

Fluorescence spectroscopy is the measurement andanalysis of various features that are related to the fluo-rescence quantum yield and/or lifetime of a biologicalmolecule. The fluorescence intensity of a biologicalmolecule is a function of its concentration, its extinc-tion coefficient (absorbing power) at the excitationwavelength, and its quantum yield at the emissionwavelength..2/ A fluorescence emission spectrum rep-resents the fluorescence intensity measured over a rangeof emission wavelengths at a fixed excitation wavelength.On the other hand, a fluorescence excitation spectrum isa plot of the fluorescence intensity at a particular emis-sion wavelength for a range of excitation wavelengths.


(a)

Flu

ores

cenc

ein

tens

ity

Emissionwavelength (nm)

Fixed excitationwavelength

(b)

Flu

ores

cenc

ein

tens

ity

Excitationwavelength (nm)

Fixed emissionwavelength

(c)

Exc

itatio

nw

avel

engt

h (n

m)

Emissionwavelength (nm)

Figure 2 Illustration of: (a) a fluorescence emission spectrum;(b) a fluorescence excitation spectrum; (c) an EEM of amolecule.

A fluorescence excitation–emission matrix (EEM) is atwo-dimensional contour plot that displays the fluores-cence intensities as a function of a range of excitation andemission wavelengths. Each contour represents pointsof equal fluorescence intensity. Figure 2 illustrates afluorescence emission spectrum (Figure 2a), a fluores-cence excitation spectrum (Figure 2b) and a fluorescenceEEM (Figure 2c).

2 FLUOROPHORES

2.1 Endogenous Fluorophores

Table 1 lists the biological molecules that exhibit endoge-nous fluorescence, along with their excitation andemission maxima..11,27/ These endogenous fluorophoresinclude amino acids, structural proteins, enzymes andco-enzymes, vitamins, lipids and porphyrins. Their exci-tation maxima lie in the range 250–450 nm (spanning theUV/VIS spectral range), whereas their emission maximalie in the range 280–700 nm (spanning the UV/VIS/NIRspectral range). Details of the molar extinction coef-ficients, fluorescence quantum yields and lifetimes areprovided elsewhere..11/ The endogenous fluorophoresthat are speculated to play a role in transformations thatoccur with carcinogenesis are the amino acids tryptophanand tyrosine,.28,29/ the structural proteins collagen andelastin,.30 – 36/ the coenzymes NADH.28,37/ and FAD,.37,38/

and porphyrins..39/

Table 1 Excitation and emission maxima of biologicalmolecules that exhibit endogenous fluorescence

Endogenous Excitation Emissionfluorophores maxima (nm) maxima (nm)

Amino acidsTryptophan 280 350Tyrosine 275 300Phenylalanine 260 280

Structural proteinsCollagen 325 400, 405Elastin 290, 325 340, 400

Enzymes and coenzymesFAD, flavins 450 535NADH 290, 351 440, 460NADPH 336 464

VitaminsVitamin A 327 510Vitamin K 335 480Vitamin D 390 480

Vitamin B6 compounds

Pyridoxine 332, 340 400Pyridoxamine 335 400Pyridoxal 330 385Pyridoxic acid 315 425Pyridoxal 50-phosphate 330 400Vitamin B12 275 305

LipidsPhospholipids 436 540, 560Lipofuscin 340–395 540, 430–460Ceroid 340–395 430–460, 540

Porphyrins 400–450 630, 690

FAD, flavin adenine dinucleotide; NADH, reduced nicotinamideadenine dinucleotide; AND(P)H, reduced nicotinamide adenine di-nucleotide phosphate.

2.1.1 Amino Acids: Tryptophan and Tyrosine

Amino acids are the basic structural units of a protein.40/

and proteins play crucial roles in virtually all of thebiological processes..40/ Three amino acids with aromaticside chains are fluorescent. At excitation wavelengthsabove 295 nm only tryptophan is fluorescent..26/ From 280to 295 nm, both tyrosine and tryptophan are fluorescent;however, energy transfer from tryptophan to tyrosineis common..26/ Below 280 nm, all three amino acids canbe excited, albeit the quantum yield of phenylalanineis relatively low compared with that of tryptophan andtyrosine..26/

2.1.2 Structural Proteins: Collagen and Elastin

Collagen, a family of fibrous proteins, is distinctivein forming insoluble fibers that have a high tensilestrength..40/ It is the major extracellular matrix com-ponent that is present to some extent in nearly all organs


and serves to hold cells together in discrete units. Elastinis the major component of elastic fibers and is found inmost connective tissues in conjunction with collagen andpolysaccharides..40/

Fluorescence has been noted from collagen.41/ andelastin..42/ Collagen fluorescence in load-bearing tissuesis associated with cross-links, hydroxylysyl pyridolineand lysyl pyridinoline..43/ Collagen fluorescence has anexcitation maximum at 325 nm and an emission maximumat 400 nm, as displayed in Figure 3..41/ The fluorescentmaterial in elastin is a tricarboxylic triamino pyridiniumderivative, which is very similar in spectral propertiesto the fluorophore in collagen..44/ The excitation andemission maxima of the fluorescence of this elastin cross-link occur at 320 and 405 nm, respectively, as shown inFigure 4..42/

2.1.3 Metabolic Coenzymes: Pyridine Nucleotides andFlavoproteins

Living organisms require a continual input of freeenergy through cellular metabolism for the performanceof mechanical work in muscle contraction and othercellular movements, the active transport of moleculesand ions, and the synthesis of macromolecules andother biomolecules from simple precursors..40/ In mostprocesses, the carrier of free energy is adenosine triphos-phate (ATP), which is derived from the oxidation offuel molecules such as carbohydrates and fatty acids.In aerobic organisms, the ultimate oxidizing agent orelectron acceptor is molecular oxygen. However, elec-trons are not transferred directly from fuel moleculesand their breakdown products to molecular oxygen.Instead, these substrates transfer electrons to special

0.1

250 300 350250 300 350 400 450

0.2

Abs

orba

nce

Wavelength (nm)

100

50

Flu

ores

cenc

e

(a) (b) (c)

Figure 3 Collagen fluorescence: (a) UV absorption spectra in0.1 N HCl (-Ð-Ð-Ð-), in 0.1 M potassium phosphate buffer, pH 7.4.- - -/, and in 0.1 N NaOH . /; (b) fluorescence excitationspectrum in 0.02 M potassium phosphate buffer, pH 7.4, withemission fixed at 400 nm; (c) fluorescence emission spectrumin 0.02 M potassium phosphate buffer, pH 7.4, with excitationfixed at 325 nm. (Reproduced by permission of Academic Press,Orlando, Florida, from Fujimoto..41/)

275 300 325 350250

0.2

0.4

0.6

0.8

1.0

1.2

1.4

1.6

Abs

orba

nce

nm(a)

460

440

400

380

360

360

340

320

nm(b)

Flu

ores

cenc

e (a

rbitr

ary

units

)

405

320

Fluorescence Activation

Figure 4 Elastin fluorescence: (a) UV absorption spectrum in0.1 M potassium phosphate buffer; (b) fluorescence emissionand excitation (activation) spectrum in 0.1 M HCl. (Reprintedfrom Deyl et al.,.42/ with permission from Elsevier Science.)

carriers called pyridine nucleotide (PN) and flavoprotein(Fp). The oxidized form of PN – nicotinamide ade-nine dinucleotide (NADC) – and the oxidized form ofFp/FAD – are the major electron acceptors in the oxida-tion of fuel molecules. After accepting the electrons, thesecarriers become reduced. The now reduced electron car-riers – NADH and reduced flavin adenine dinucleotide(FADH2) – then transfer their electrons to molecularoxygen by means of the electron transport chain in theinner membrane of the mitochondria within the cell. Asa result, molecular oxygen oxidizes them to NADC andFAD. The ultimate flow of these electrons to molecularoxygen drives the synthesis of ATP.

The fluorescent forms of these electron carriers are thereduced form of PN and the oxidized form of Fp, whichemit fluorescence when excited with UV and blue lightrespectively. Figure 5 displays the fluorescence excitationand emission spectra of Fp and PN from isolated pigeonheart mitochondria measured at �196 °C..45/ Figure 5(a)and (b) displays the oxidized (Ox) and reduced (Red)Fp excitation and emission spectra, respectively, andFigure 5(c) and (d) shows the oxidized (Ox) and reduced(Red) PN excitation and emission spectra, respectively.Clearly, the excitation and emission of Fp is maximalwhen it is oxidized and minimal when it is reduced. Theconverse is true for PN. The ratio of the concentration ofoxidized and reduced electron carriers (or, synonymously,the ratio of the fluorescence of Fp and PN) gives a measureof the cellular metabolic rate..45/ In particular, a reduced


(a)

Rel

ativ

e in

tens

ity (

a.u.

) Fp

400 450 500

100

0

Ox

λ (nm)

520 nm emission

Red

(b)

Rel

ativ

e in

tens

ity (

a.u.

)

Fp

500 550 600

100

0

Ox

λ (nm)

460 nm excitation

Red

(c)

Rel

ativ

e in

tens

ity (

a.u.

) PN

320 360 400

100

0

Red

λ (nm)

460 nm emission

Ox

(d)

Rel

ativ

e in

tens

ity (

a.u.

)

PN

400 450 500

100

0

Red

λ (nm)

366 nm excitation

Ox

Figure 5 (a, b) The oxidized (Ox) and reduced (Red) Fpexcitation and emission spectra, respectively. (c, d) The oxidized(Ox) and reduced (red) PN excitation and emission spectra,respectively. The measurements were made from pigeon heartmitochondria at �196 °C. (Reproduced by permission fromChance et al..45/)

ratio is indicative of a high phosphate potential (state 4),whereas an oxidized ratio indicates a low phosphatepotential (states 2 and 3).

2.1.4 Porphyrins

Porphyrin is a precursor in heme biosynthesis..40/ Itleads to the synthesis of protoporphyrin IX (PpIX),which acquires an atom to form heme. d-Aminolevulinatesynthase, the enzyme catalyzing the committed step inthis pathway, is feedback inhibited by heme. Porphyrinsproduce a red fluorescence, with peaks at 630 and 690 nm,when excited in the blue spectral region between 400 and450 nm..46/

2.1.5 Fluorescence Microscopy and Spectroscopy ofMicrostructures in Cultured Cells and TissueSections

Fluorescence microscopy and spectroscopy of fluores-cent micro-structures in cultured cells and optically thin,unstained, frozen tissue sections have indicated that dif-ferences in the fluorescence emission spectra of normal,

precancerous and cancerous tissues may be attributedto differences associated with several endogenous fluo-rophores within the various sublayers of these tissues.Pradhan et al..28/ suggested that there is an increasein NADH and tryptophan, whereas Anidjar et al..38/

reported that there is probably a decrease in FAD ascells progress from a normal to cancerous state. Romeret al..30/ and Fairman et al..32/ observed a decreased flu-orescence intensity from the collagens in precancerouscolon tissue sections relative to that in normal colontissue sections. Bottiroli et al..34/ observed a red fluores-cence in some parts of a tissue section that was from acancerous colon. The red fluorescence that was observedby Bottiroli et al..34/ has been observed previously byseveral groups..37,47 – 51/

2.2 Exogenous Fluorophores

Most exogenous fluorophores currently being eval-uated as contrast agents for precancer and cancerdetection include photosensitizers, developed originallyfor photodynamic therapy (PDT)..12/ These includehematoporphyrin derivative (HpD), pheophorbide-a,meso-tetra-(hydroxyphenyl)-chlorin (MTHPC), benzo-porphyrin derivative (BPD), tin etiopurpurin (SnET2),hypercin and pthalocyanine..12/ These compounds havestrong absorption characteristics in the blue spectralregion or up to the highest Q-band at 635 nm, resultingin fluorescence between 625 and 675 nm..12/ These agentsare generally administered systemically and the differ-ence in accumulation of these exogenous fluorophores intumor tissue relative to their normal counterpart appearsto correlate with the differences in the vascularity of thetwo tissue types..12/ Lutetium texaphyrin is another pho-tosensitizing agent that has absorption and fluorescencecharacteristics in the NIR spectral region..12/ This agentseems to be selectively phototoxic to tumors via apoptosisrather than necrosis (which results from vascular stasis),suggesting that this agent localizes in the cell..12/

During the past few years an alternative concepthas been introduced, which is based on the initialobservations of Ghadially et al.,.51/ who suggested usingthe photosensitizing precursor d-aminolevulinic acid(5-ALA) to induce PpIX fluorescence in tumors. PpIXis characterized by an absorption maximum at 405 nmand fluorescence maxima at 630 and 700 nm..52/ Thecompound 5-ALA is a natural precursor of PpIX inthe biosynthetic pathway for heme..52/ Normally, thesynthesis of heme regulates the synthesis of PpIX throughfeedback control. However, the administration of 5-ALAbypasses this feedback and induces the accumulation ofPpIX in tumor tissue. It has been suggested that deficiencyin ferrochelatase (the enzyme required for conversion ofPpIX to heme) in tumors results in accumulation of PpIX


in these tissues relative to normal tissue..12/ There areseveral advantages associated with using 5-ALA as acontrast agent. Both 5-ALA and PpIX are substancesnaturally present in the body, making the toxicity issueless critical. Furthermore, the drug can be administeredconveniently both orally and topically..12/

The use of photosensitizers as purely contrast agentshas raised potential safety and toxicity concerns. Hence,recent work has examined nonphotosensitizing agentswith similar absorption and fluorescence characteristicsto photosensitizers for their potential to serve as con-trast agents. Several exogenous fluorophores, includingNile blue and its derivatives and the caretenoporphyrins,have been shown to fulfil this criterion..12/ Nonphoto-toxic or weakly phototoxic exogenous fluorophores thatare in routine clinical use currently are fluorescin (flu-orescin angiography) and indocyanine green (cardiacoutput measurements and liver assessment), which haveabsorption and fluorescence characteristics in the VIS andNIR regions, respectively..11/ In clinical use, these drugsare injected into the vasculature, through which they areultimately cleared.

3 FLUORESCENCE SPECTROSCOPY OFTURBID MEDIA

3.1 Fluorescence Spectroscopy of an Optically Dilute,Homogeneous Medium

Using the Beer–Lambert law,.2/ one can express thefluorescence intensity of an optically dilute, homogeneousmedium (sum of the optical densities at the excitationand emission wavelengths. Everywhere is less thanunity).26/ as a linear function of the concentrationof the fluorophores in that medium. Specifically, thefluorescence intensity, F.lx, lm/, at particular excitation(lx) and emission (lm) wavelengths due to k fluorophoresis defined according to Equation (2):

F.lx, lm/ D P0.lx/∫ L

0dz �

N∑kD1

µak.lx/fk.lm/ .2/

and Equation (3) gives the definition of µa.lx/, theabsorption coefficient of the fluorophore at the excitationwavelength:

µa.lx/ D 2.303e.lx/C .3/

In Equations (2) and (3) f.lm/ is the quantum yield ofthe fluorophore at the emission wavelength, P0.lx/ is thepower of the intensity at the excitation wavelength, L isthe path length and� is the detector collection efficiency.The absorption coefficient µa.lx/ is a linear functionof e.lx/ and C, which are the extinction coefficients at

excitation wavelength and concentration, respectively, ofthe fluorophore. Although fluorescence spectroscopy ofoptically dilute, homogeneous media is well understood,fluorescence spectroscopy of turbid media, such ashuman tissue, is complicated by its highly absorbing andscattering properties..53/

3.2 Fluorescence Spectroscopy of Turbid Media such asTissue

Fluorescence spectroscopy of turbid media such as tissuedepends on one or more of the following. Specifi-cally, it depends on the concentration and distributionof fluorophore(s) present in the tissue as well as thebiochemical/biophysical environment, which may alterthe quantum yield and lifetime of the fluorophore(s).For example, epithelial tissues generally have two pri-mary sublayers – a surface epithelium and an underlyingstroma or submucosa; the specific fluorophores, as wellas their concentration and distribution, can vary signif-icantly between these two tissue layers. Fluorescencespectroscopy of turbid media such as tissue also dependson the absorption and scattering that result from the con-centration and distribution of nonfluorescent absorbersand scatterers, respectively, within the different sublayersof the tissue.

The effect of the aforementioned variables on fluores-cence spectroscopy of tissue is wavelength dependent.First, the endogenous fluorophores that have absorp-tion bands that lie in the same wavelength range as theexcitation light will be excited and hence will emit flu-orescence. The absorption and scattering properties ofthe tissue will affect light at both of these excitation andemission wavelengths. Therefore only those fluorophorescontained in the tissue layers to which the excitation lightpenetrates and from which the emitted light can escapethe tissue surface will produce measurable fluorescence.Elastic scattering events in tissue are caused by randomspatial variations in the density, refractive index anddielectric constants of extracellular, cellular and subcellu-lar components..54/ Tissue scattering generally decreasesmonotonically with increasing wavelength over the UV,VIS and NIR spectral regions..11/ Tissues are generallyforward scattering (mean anisotropy factor is between0.97 and 0.98), with scattering coefficients ranging from10 to 1000 cm�1 from the NIR to the UV spectral range..11/

Absorption in tissue in the UV, VIS and NIR regions isprimarily attributed to hemoglobin..55/ Although absorp-tion in tissue is strongly wavelength dependent, it tends todecrease generally with increasing wavelengths..11/ Typ-ically, tissue absorption coefficients range from 0.1 to10 000 cm�1 from the NIR to the UV spectral range.Consequently, the penetration depth of light, which isprimarily a function of the tissue absorption properties,


decreases from several centimeters to a few hundredmicrons from the NIR to the UV..11/ For example, inthe UV spectral region, the penetration depth of light intissue is approximately 225 µm at 337 nm..3/

Hemoglobin, which is contained in red blood cells,serves as the oxygen carrier in blood and also playsa vital role in the transport of carbon dioxide and

60

120

180

5

10

15

400 450 500 550 600 650

ε×10

−3 (m

ol l−1

cm

−1)

ε×10

−3 (m

ol l−1

cm

−1)

Wavelength (nm)(a)

20 000

10 000

5000

1000

500

100

50

10500 550 600 650 700 750 800 850 900

Mol

ecul

ar e

xtin

ctio

n co

effic

ient

HbO2 Hb

λ (µm)(b)

Figure 6 Absorption spectra of oxygenated and deoxygenatedhemoglobin over: (a) UV/VIS spectral region (solid line,HbO2; dashed line, Hb; dotted line, Soret band); (b) NIRspectral region. (Reproduced by permission of Academic Press,Orlando, Florida, from Brown..55/)

hydrogen ions..40/ The capacity of hemoglobin to bindoxygen depends on the presence of a nonpolypeptide unit,namely a heme group. The heme consists of an organicpart, a protoporphyrin ring and an iron atom. Figure 6displays the absorption spectra of oxygenated anddeoxygenated hemoglobin over the UV/VIS (Figure 6a)and NIR (Figure 6b) spectral regions..55/ In Figure 6(a),the absorption spectra are characterized by the Soret bandat 400–450 nm, the bands at 540 (a band) and 569 nm (bband) of oxygenated hemoglobin and the 557 nm band ofdeoxygenated hemoglobin. In Figure 6(b), the absorptionspectra are characterized by the decreasing absorption ofdeoxygenated hemoglobin and the increasing absorptionof oxygenated hemoglobin as a function of increasingwavelength.

The effect of hemoglobin absorption is noticeablewhen comparing fluorescence emission spectra measuredfrom an optically thick arterial tissue sample (250 µmthickness) to that measured from a corresponding,optically thin tissue section (4 µm thickness) using thesame illumination and collection geometry, as shownin Figure 7..56/ Evaluation of Figure 7 indicates that thefluorescence intensity of the optically thick tissue sampleis significantly reduced and, furthermore, its line shape ischaracterized by a valley at 420 nm, which correspondsto the Soret absorption band of hemoglobin..55/ Thefluorescence emission spectrum of the optically thin tissuesection may be attributed to collagen fluorescence.

The illumination and collection geometry of theexcitation and the emitted light, respectively, can alsoaffect fluorescence measurements from tissue with respectto both the intensity and line shape..54,57,58/ This maybe attributed to the fact that although fluorescence isgenerated isotropically from the fluorophores within the

3600.00E + 00

5.00E + 04

1.00E + 05

1.50E + 05

2.00E + 05

2.50E + 05

3.00E + 05

410 460 510 560 610

Flu

ores

cenc

e in

tens

ity (

a.u.

)

Wavelength (nm)

TurbidDilute

Figure 7 Fluorescence emission spectra at 340 nm excitation,measured from an optically thick, arterial tissue sample (250 µmthickness) to that measured from a corresponding, opticallythin, tissue section (4 µm thickness) using the same excitationand collection geometry. (Reproduced by permission of theOptical Society of America, from Durkin et al..56/)


biological medium, the fluorescence emitted from thesurface of the medium may range from isotropic toanisotropic, depending on whether the medium is highlyabsorbing, dilute or turbid..54/ Monte-Carlo simulationshave been used extensively to simulate light distributionin turbid media to explore the effect of absorption andscattering on the fluorescence emitted from the surface oftissues using finite excitation beam profiles and complexexcitation and emission geometries..57/ The results ofthese simulations indicate several important findings.

The profile of the excitation beam greatly affects thedistribution of the excitation light in the tissue and thusis an important factor. For a Gaussian beam profile, thefluorescence distribution at a specific emission wavelengthon the surface of the tissue is peaked and narrow,whereas for a uniform profile it is wider and less peaked.Furthermore, the effect of absorption, particularly dueto hemoglobin, on the measured fluorescence increasesas the collection fiber is moved further away from theillumination fiber on the surface of the tissue, resulting indecreased fluorescence intensity. In addition, the effectof this absorption is wavelength dependent, suggestingthat it will affect the line shape of the fluorescence aswell. This is evident when comparing the fluorescenceemitted at two wavelengths at which hemoglobin hasdifferent absorption characteristics. As the distance ofthe collection fiber increases relative to the illuminationfiber, the ratio of the fluorescence intensity at the twoemission wavelengths deviates from unity. This suggeststhat collecting the emitted light from the tissue sitedirectly illuminated by the excitation light will minimizethe distortion due to hemoglobin absorption, albeit thiswill reduce the sampling depth for the fluorescencemeasurements.

The effect of the illumination and collection geometryon fluorescence measurements from tissue is illustratedin Figure 8, which displays the fluorescence emissionspectra at 476 nm excitation of normal human cadaver

aorta (N) and of one with atheromatous plaque (P)..58/ InFigure 8(a) fluorescence was collected from the tissuesite directly illuminated by the excitation light, andin Figure 8(b) the fluorescence was collected from acircular area around the directly illuminated area. Boththe intensity and line shape of the fluorescence areaffected by the probe geometry. For the normal aorta(Figure 8a) the ratio of the fluorescence intensity at 600and 580 nm is 1.27, whereas it is 1.72 in Figure 8(b).For the atheromatous tissue, the ratio is 0.91 for bothcollection areas.

3.3 Deconvolution of Absorption and Scattering fromTissue Fluorescence Emission Spectra

Fluorescence excitation and emission from turbid mediasuch as tissues consist of three components: thedistribution of the excitation light in the tissue, whichis a function of the absorption and scattering coefficientsof the medium at that wavelength; the fluorescence ofisotropically radiating point sources located at differ-ent depths within the tissue, which is determined bythe fluorescence quantum yield of the fluorophore andthe excitation intensity at that depth; and the total flu-orescence escaping the surface of the tissue, which is afunction of the absorption and scattering properties of themedium at that wavelength. Quantification of the concen-tration of fluorophores within tissue in principle involvesdeconvoluting the absorption and scattering propertiesof the tissue from the measured fluorescence emissionspectrum and quantifying the identity and distributionof the fluorophores that contribute to the deconvolvedspectrum. The former requires the development of atransfer function based on the measurements of tissueoptical properties, i.e. the absorption and scattering coef-ficients and the anisotropy parameter, whereas the latterrequires knowledge of the identity and distribution of thefluorophores within the tissue.

2 mm500 550 600 650 500 550 600 650

Wavelength (nm) Wavelength (nm)

P

N

N

P

1 mm

F(λ) = F(λ,r) 2πr dr ∫0

0.5 mm F(λ) = F(λ,r) 2πr dr ∫1 mm

0.5 mm

(a) (b)

Figure 8 Fluorescence emission spectra at 476 nm excitation of normal human cadaver aorta (N) and one with atheromatousplaque (P): (a) the fluorescence was measured from the tissue site directly illuminated by the excitation light; (b) the fluorescencewas measured from a circular area around the directly illuminated area. (Reproduced by permission of the Optical Society ofAmerica, from Keijzer et al..58/)


For a turbid medium such as tissue that contains kfluorophores, the fluorescence intensity can be rewrittento include a transfer function (TF) that describes theattenuation of the excited and emitted light due to theabsorption and scattering properties of the tissue at thesewavelengths..54/ This is defined in Equation (4):

F.lx, lm/ D P0.lx/∫ 1

0dz

N∑kD1

µak.lx, z/fk.lm, z/

ð TF.lx, lm, z/� .4/

where z refers to the depth within the medium. Here, thetransfer function for a particular depth can be determinedfrom the tissue absorption coefficient, the scatteringcoefficient and the anisotropy factor. The wavelength-dependent absorption coefficient (µa) denotes the proba-bility of photon absorption per unit path length, whereasthe scattering coefficient (µs) denotes the probability ofphoton scattering per unit path length. The anisotropyfactor (g) denotes the cosine of the average scatteringangle. The reduced or isotropic scattering coefficient (µ0s)is the product of the scattering coefficient (µs) and oneminus the anisotropy parameter (1� g). Models of lighttransport.59 – 65/ can be used to deduce the absorptionand scattering coefficients and the anisotropy parameterneeded to compute the transfer function in Equation (4).

3.3.1 Models of Light Transport

Most models of light transport used to quantify tissueoptical properties to date have been either numerical orapproximate analytical solutions of the neutral particletransport equation..59,60/ Because of the inhomogeneity oftissues, the solution of Maxwell’s equations, which mightotherwise be used to model light propagation in tissues,is generally not feasible. However, if polarization anddiffraction effects are ignored, the transport of photonsthrough random media may be modeled as neutralparticle transport. The neutral particle transport theoryis heuristic and is based on a statistical approximation ofphoton transport in a multiplescattering medium. Withinthis framework, the propagation of light is described interms of the transport of discrete photons, which maybe either scattered or absorbed. The transport equation(Equation 5) is defined as:

[email protected], �̂/

@tC �̂ ž rf.Er, �̂/C [µa.Er/C µs.Er/]f.Er, �̂/

D µs.Er/∫

4pd�̂0f .�̂0 ! �̂/f.Er, �̂0/C s.Er, �̂/ .5/

where f.Er, �̂/ is the local angular flux of photons atposition Er, angle �̂ and time t in a scattering medium. Theabsorption coefficient µa.Er/ describes photon absorption

at position Er, and the scattering coefficient µs.Er/ describesphoton scattering at position Er. The term f .�̂0 ! �̂/ is thescattering phase function, which describes the probabilityof a photon scattered from direction �̂0 into direction �̂.Term S.Er, �̂/ is a source term for photons at position Erand angle �̂. Equation (5) can be solved numerically orapproximated analytically to determine the position- andangle-dependent flux of photons in a medium describedby an arbitrary absorption, scattering and phase functiondistribution.

Approximate analytical models include the followingapproximations. The simplest analytical model is theBeer–Lambert law,.2/ which assumes that the incidentlight on the medium attenuates exponentially in thedirection of propagation within that medium. This isvalid for a medium that does not scatter light, but issufficiently absorbing that it can no longer be dilute, whichis generally not true for tissues. The Kubelka–Munkapproximation.61/ describes the time-independent, diffusereflectance and transmittance for light incident on a tissueslab in terms of its spatially uniform absorption andisotropic scattering coefficients. The radiative transportequation also can be solved using the adding–doublingmethod.62/ for appropriate boundary conditions, and itssolution can be compared with measurements of time-independent, diffusely reflected and transmitted light toestimate the absorption and scattering, as well as theanisotropy factors. Models based on Kubelka–Munkand the adding–doubling method are appropriate forthe determination of tissue absorption and scatteringcoefficients in a wavelength region in which neitherabsorption nor scattering in tissue is dominant (UV/VISspectral regions).

Within the diffusion approximation,.63/ which is validfar from sources and when the scattering coefficient isseveral orders of magnitude greater than the absorptioncoefficient, the spatial and temporal distribution of thediffusely reflected or transmitted light from tissue isrelated to the absorption and scattering coefficients andanisotropy factor. Diffusion theory is valid for caseswhere the absorption is low and scattering is high intissues, such that the light is multiply scattered andtravels through various optical path lengths before beingdetected (as in the NIR spectral region). When theprobability of absorption in tissue becomes significant(as in the UV/VIS spectral regions), re-emitted light isminimally or moderately scattered and diffusion theory isno longer valid.

Numerical solutions to the transport equation includethe method of discrete ordinates.64/ and the Monte-Carlotechnique..65/ These techniques can be used to model lighttransport in tissue over the entire UV/VIS/NIR spectralregions. The Monte-Carlo technique, which has been the


more popular of these numerical models, tracks individ-ual photon trajectories, which are computer-simulated tocalculate absorption and scattering coefficients (inversecalculation) or space irradiance distributions (forward cal-culation) for any tissue, tissue geometry and wavelengthregion. However, because tens of millions of photon tra-jectories must be tracked for statistical simulation of lighttransport in tissue, computations are time intensive.

The aforementioned numerical techniques can be usedto compute the absorption and scattering coefficients andanisotropy parameter from measurements of diffuselyreflected and transmitted light from tissue. Additionally,if the optical properties are known at the specific exci-tation and emission wavelengths and the fluorophore(s)quantum yield and distributions are provided, fluores-cence excitation, emission and escape from the differentdepths within the tissue can be computed. Hence, thesenumerical techniques can be used to compute numericallythe transfer function in Equation (4).

3.4 Turbid Tissue-simulating Phantoms forFluorescence Spectroscopy of Tissue

In order to evaluate the effect of absorption and scatteringand the illumination and collection geometry experimen-tally on fluorescence measurements from tissue, it isuseful to employ tissue phantoms that simulate the opticalproperties and fluorescent properties of these biologicalmedia.

Durkin et al..66/ developed a liquid tissue phantomthat simulates the optical and fluorescence propertiesof tissues over the range 350–650 nm, with absorptioncoefficients in the range 25–5 cm�1, scattering coef-ficients, in the range 400–200 cm�1 and fluorophoresin the micromolar concentration range. Phantom fluo-rophores included FAD and rhodamine B. Absorptionwas controlled by adjusting the hemoglobin concentra-tion. Polystyrene spheres, which have a smoothly varyingscattering coefficient as a function of wavelength, wereused as the scatterers. Sample inhomogeneities were sim-ulated by preparing the phantom in a gelatin substrate.These phantoms could be made dilute, absorbing and/orturbid.

Wagnieres et al..67/ also developed tissue phantomsthat have the optical characteristics of biological tissuesin the wavelength range 400–650 nm, with absorptioncoefficients in the range 21–0.7 cm�1 and scatteringcoefficients in the range 680–220 cm�1. The phantomsare made up of agarose dissolved in water to providea transparent matrix, which is then loaded with silicondioxide, Intralipid, ink, blood, azide, penicillin, bovineserum and fluorophores. The silicon dioxide and Intralipidparticles are responsible for scattering, whereas theink and blood are absorbers. The penicillin and azide

are used to ensure conservation of the phantoms at4 °C. The serum, and fluorophores such as Coumarin 30,produce fluorescence. The mechanical properties of thesegelatinous phantoms render them easily moldable so thatcomplex structures and shapes simulating layered or otherinhomogeneous structures containing various amounts ofabsorber, scatterer and fluorophore can be developed.

4 INSTRUMENTATION

In order to perform quantitative fluorescence spec-troscopy of tissue, the fluorescence emission spectra andthe reflectance and/or the diffuse reflectance and trans-mittance need to be measured. The latter are needed iftissue optical properties are to be quantified. However,because intact issues are semi-infinite media, only fluo-rescence emission and reflectance spectra are measuredfrom these biological systems.

A schematic of the basic components of an instrumentused for fluorescence and/or reflectance spectroscopyof tissues is shown in Figure 9. It consists of a lightsource, a flexible, conduit that contains optical fibersfor the illumination and collection of light, a dispersingelement that separates the emitted light into its respectivewavelengths, and a detector that measures the intensityat these wavelength(s). In this scenario, fluorescenceemission and reflectance are measured in a re-emissiongeometry in which the illumination and collection areperformed on the same surface of the tissue. The varioustypes of instruments employed for these measurementsessentially have the same basic components.

4.1 Light Sources

Lasers, which have a very narrow spectral output,are generally used as monochromatic excitation lightsources for fluorescence spectroscopy, whereas lampswith a broad spectral output are more appropriate for

Lightsource

Dispersingelement

Detector

Conduit for illumination andcollection of light

Biological system

Figure 9 Schematic drawing of a fiber-optic-based instrumentfor fluorescence and/or reflectance spectroscopy.


reflectance spectroscopy. Lamps also can be used as quasi-monochromatic excitation light sources for fluorescencespectroscopy when coupled with a monochromator ornarrow-bandpass filter (see section 4.3). The need forportable light sources for clinical applications has limitedthe use of monochromatic light sources in the UV/VISspectral range to the nitrogen pumped-dye laser (funda-mental wavelength is 337 nm) and the helium–cadmiumlaser (wavelengths are 325 and 442 nm). Additionally, adye module coupled to the nitrogen laser enables thegeneration of VIS laser wavelengths through the use ofappropriate chemical dyes in a resonant cavity. Otherlasers that have been used for fluorescence spectroscopyinclude the argon ion (UV and blue), helium neon (red)and krypton ion (blue) lasers. Portable polychromaticsources include the mercury and xenon arc lamps. The pri-mary criteria for the selection of light sources, in additionto portability, are output power, bandwidth, wavelengthtunability, coupling efficiency (into optical fibers) and theneed for pulsed versus continuous wave light sources.

The power from a portable nitrogen laser at 337 nmis ¾300 µJ per pulse. At a maximal repetition rate of30 Hz its average power is ¾9 mW. Because the nominalconversion efficiency of dyes used to generate VISwavelengths is approximately 10%, the average powerin this case would be ¾1 mW. The output power fromthe helium–cadmium laser ranges from 5 mW at 325 nmto 15 mW at 442 nm. Commercially available mercuryand xenon arc lamps can be operated over a range of50–1000 W. Owing to excess generation of heat in the IR,clinical applications employing these lamps have beenrestricted to the use of power supplies between 75 and300 W, with a liquid filter for IR rejection. Note thatthe output powers available for tissue illumination aresignificantly lower than the output powers specified, dueto coupling losses from the light source into the opticalfibers and/or coupling optics.

Although the UV/VIS lasers afford a very narrowspectral band (1–2 nm), the bandwidth of illuminationfrom a lamp is determined by the use of narrow-bandpassfilters or a monochromator (see section 4.3), which isused for wavelength dispersion. In general a 10–20 nmbandpass for the light source is sufficient, because tissuefluorescence emission spectra are broad with a 40–60 nmfull width at half-maximum (fwhm).

The advantages and disadvantages of using a lampversus a laser for fluorescence spectroscopy shouldbe considered. An advantage of using a mercury orxenon lamp over a laser is that it provides wavelengthtunability over the UV/VIS spectral range, as shownin Figure 10..68/ However, the disadvantage of using alamp over a laser is that the coupling efficiency intooptical fibers can be reduced by a factor of 20 comparedto laser coupling,.69/ due to the relatively large focal

6.0

4.0

2.0

6.0

4.0

2.0

250 275 300 350 400 500 700

mol

s−1

ste

r−1 ×

107

(left

hand

sec

tion)

mol

s−1

ste

r−1 ×

106

(rig

ht h

and

sect

ion)

(a)

1.5

1.0

0.5

1.5

1.0

0.5

250 300 400 500 800350

mol

s−1

ste

r−1 ×

107

(mid

dle

sect

ion)

mol

s−1

ste

r−1 ×

106

(rig

ht h

and

sect

ion)

λ (nm)(b)

Figure 10 Typical radiant intensity and shape of the emissionspectra of: (a) mercury lamp; (b) xenon arc lamp. (Reproducedby permission of Academic Press, Orlando, Florida, from Kohenet al..68/)

spot diameter of arc lamps. Finally, lasers with veryshort pulse duration (of the order of nanoseconds)are necessary when the tissue needs to be illuminatedwith pulsed excitation light for gated detection (whichprovides effective rejection of ambient light duringfluorescence measurements) and for fluorescence lifetimemeasurements.

4.2 Illumination and Collection of Light

With respect to illumination and collection of light fromtissue, two different approaches may be considered: thecontact approach, where fiber-optic probes are placeddirectly in contact with the tissue surface; and thenoncontact approach, where a series of lenses are usedto project the light onto the tissue surface and collect itin a similar manner. With the contact approach, variablepressure on the tissue may distort the signal. However,with the noncontact approach the signal strength will vary


with the source–tissue and tissue–detector distances.In general, the contact approach is used for steady-state and time-resolved fluorescence and reflectancespectroscopy from small tissue volumes, whereas thenoncontact approach is more suited for imaging largertissue areas.

For the fiber-optic-based approach, typically multi-mode, step-index fibers with a core diameter rangingfrom 100 µm to 1 mm and f numbers (see section 4.3) of2–4 are used for UV/VIS, fluorescence and reflectancespectroscopy. Fused-silica optical fibers are used forUV-excited fluorescence spectroscopy because stan-dard silica fibers generate fluorescence with UVexcitation.

The fluorescence emission spectra of turbid media suchas tissue are extremely dependent on the illumination andcollection geometry due to the interaction of absorptionand fluorescence..57/ As shown elsewhere,.57,58/ designof a fiber-optic probe configuration that reduces thedistortion of the fluorescence emission spectrum due totissue absorption consists of a geometry in which thetissue fluorescence is collected only from that surfacedirectly illuminated by the excitation light. Richards-Kortum et al..57/ first designed optical fiber probeswith illumination and collection geometries that enablemaximal overlap between excitation and emission areason the tissue surface.

In general, reflectance spectroscopy refers to thedetection of both the diffuse (multiply scattered) andspecular (surface) components of the reflectance. Whenmeasuring the diffuse reflectance from tissues, thephotons that are detected generally undergo multiplescattering events and the photon path lengths are muchgreater than the geometric separation between the sourceand detector. Consequently, through this measurement,the probe is able to sample features in the tissuevolume. Because specular reflection is due to refractiveindex mismatch and is generally a 2% reflection of theincident light in the case of an air/tissue interface, itneeds to be minimized in attempting to make diffusereflectance measurements. In order to minimize specularreflection, geometries have been employed in whichthe source and detector fiber are placed adjacent toeach other, so that the specularly reflected light is notcoupled to the detector fiber..9/ However, this geometryprecludes the ability to collect diffuse reflectance fromthe same site that is illuminated. This presents twoproblems: there is attenuation of the signal; and iffluorescence measurements are made from the samesite of illumination there will be a lack of congruencebetween the fluorescence and reflectance measurements.One way to minimize specular reflection in a case wherethe diffuse reflectance is collected from the same site thatis illuminated is to have the distal tip of the collection

fiber angled at 17° from the direction normal to the opticalfiber axis, such that the specular reflection is not coupledbut the diffuse reflectance is..69/

4.3 Monochromators and Spectrographs

Light can be dispersed spectrally using a monochromatoror a spectrograph, which are both dispersing components.A monochromator presents one wavelength or bandpassat a time of the input light from its exit slit, whereas a spec-trograph presents a range of wavelengths simultaneouslyat the exit focal plane. Monochromators can be used asfilters in conjunction with arc lamps to produce a series ofmonochromatic outputs for sample illumination (if onlya few wavelengths are needed, narrow-bandpass filtersmay be more appropriate) or can be used to dispersethe emitted light into its respective wavelengths, eachof which can be detected serially using a single-channeldetector. Spectrographs can used to disperse the emittedlight into its respective wavelengths simultaneously formultichannel detection.

The key components of a monochromator/spectro-graph are: an entrance slit; a collimating lens; a grating(dispersing unit) for wavelength selection; a focusing lens;and an exit slit (see Figure 11). Grooves in diffractiongratings are manufactured either classically with the use ofa ruling engine (ruled gratings) or holographically with theuse of interference fringes generated at the intersectionof two laser beams (holographic gratings). Groovedensities generally range from 50 to 3600 grooves mm�1.Holographic gratings have a higher groove density thanruled gratings.

Parameters that are relevant to the selection of amonochromator/spectrograph are the reciprocal lineardispersion (mm nm�1), transmission efficiency, f numberand stray light rejection. The reciprocal linear dispersion(which is inversely related to the grating groove den-sity, diffraction order and focal length of the focusingelement), when multiplied by the entrance slit width,provides the spectral resolution. For fluorescence and

Exit slit

Grating

CondenserM2

CollimatorM1

Entrance slit

S L

��

��

(

Figure 11 Schematic drawing of a monochromator that showsthe arrangement of the light source (S), the condensing lens(L), the entrance slit, the mirror optics (M1 and M2), the gratingand the exit slit. (Reproduced by permission of Academic Press,Orlando, Florida, from Kohen et al..68/)


reflectance spectroscopy, generally a spectral resolutionof 10 nm is sufficient because these spectra are broadband and hence lack structure within this bandpass.Transmission efficiencies of both holographic and ruledgratings are comparable (¾40%) in the VIS spectralregion, and the efficiency of a particular grating is max-imal at the wavelength at which it is blazed. Collectionefficiency is reflected by the f number, which is definedas the ratio of the focal length and diameter of thecollimating element. Lower f numbers are usually asso-ciated with a higher light gathering power or throughput.However, higher f -number spectrographs with longerfocal-length-collimating elements are generally used toproject a wide, flat field ideal for multichannel imagingdetectors. For the applications indicated here, generallyf numbers that match those of standard optical fibers(¾2–4) are selected in order to maximize coupling effi-ciency. Stray light in monochromators/spectrographs isdue to diffusion of light by optical components, includ-ing gratings and mirrors. Stray light is reduced by usingholographic gratings, which usually give an order of mag-nitude better stray light rejection than classically ruledgratings. Typical stray light rejection for UV/VIS, fluo-rescence and reflectance spectroscopy should be specifiedto five orders of magnitude lower than the signal. Other-wise, long-pass filters with an optical density of at least5.0 at the wavelength of excitation have to be employedto block the excitation light from the detector.

4.4 Detectors

The important considerations in choosing a detector arethe type of measurements being made, i.e. single wave-length versus multiwavelength and single pixel (smallregion of the tissue surface) versus multipixel (relativelylarger tissue area). Fluorescence and reflectance mea-surements from single pixels can be made either usinga single-channel or multichannel detector. If measure-ments at only one or several wavelengths are being made,a single-channel photoemissive tube – called a photomul-tiplier tube (PMT) – or a semiconductor-based avalanchephotodiode (APD) with a bandpass filter can be used.For fluorescence or reflectance spectroscopy, a spectro-graph coupled to a multichannel, linear photodiode array(PDA) is appropriate. Fluorescence and reflectance spec-troscopy also can be performed using a monochromatorcoupled to a PMT, albeit this substantially increases themeasurement time. In the case of measurements frommultiple pixels on the tissue surface, a two-dimensionalcharge-coupled device (CCD) camera may be employed.In order to reduce or eliminate the detection of ambientlight during fluorescence spectroscopy, a detector with anintensifier for fast gating (several nanoseconds) must beused in conjunction with a pulsed excitation light source.

With respect to general detector performance, otherconsiderations include quantum efficiency and sourcesof noise. For the detector, the quantum efficiency isdefined as the ratio of induced current to the induced flux(often measured in electrons per photon). The quantumefficiency depends on the wavelength of light used, thematerial type and shape and other physical parameters.Figure 12 displays the quantum efficiency of a PMT andPDA/CCD..70/

The noise components that we must consider whendetecting light are as follows:

ž Shot noise, which is due to random statisticalfluctuations of the incident light. Shot noise increaseswith the square root of the signal.

ž Dark signal, which comes from the random gen-eration of electrons. Dark signal is temperaturedependent and thus can be reduced by cooling thedetector.

ž Readout noise, which comes from the electronicprocess of reading the signal from the detector.

The dark flux for a PMT lies in the range 2–50 e� s�1.The readout noise is essentially zero because the pulse-height discrimination effectively avoids quantizationnoise. Although the un-intensified PDAs have beenapplied where the light levels are high, as in the case ofreflectance spectroscopy, intensified PDAs are more suit-able for fluorescence spectroscopy in which light levels areseveral orders of magnitude lower. In an intensified PDA,a microchannel plate intensifier is used to amplify the pho-toelectrons by a factor of 3000. Typical dark (at �20 °C)and readout noises are 40 counts s�1 and 3 e�, respectively(manufacturer specification is 1 count per e�). The darkcurrent of the CCD cameras range from 12 e� pixel�1 s�1

at �40 °C to 0 e� pixel�1 s�1 when cooled to �110 °C.Their readout noise ranges from 12 to 4 e�.

4540353025201510

50250 400

22 00018 000 16 000 14 000 12 000 10 000

600 800 1000

PMT (typ.)

CCD (Si)

% Q

uant

um e

ffici

ency

ν (cm−1)

λ (nm)

Figure 12 Quantum efficiency of a PMT and PDA/CCD.


4.5 Signal-to-noise Ratio Analysis of an InstrumentUsed for Fluorescence Spectroscopy of Tissue

To establish the performance of an instrument forfluorescence spectroscopy of tissue (fluorescence ratherthan reflectance measurements are limiting in signalto noise), Zangaro et al..69/ did a signal-to-noise ratioanalysis using typical components. They used a nitrogenpumped-dye laser coupled to a fiber-optic probe forillumination and collection, and a spectrograph coupledto an intensified PDA for detection. Their calculationis as follows. The output pulse energy of the nitrogenlaser is ¾300 µJ. If the dye module is used, the nominalconversion efficiency is around 10%, but with surfacereflection losses it becomes 25 µJ per pulse. Theycalculated losses of 40% in coupling this light to a 200-µmcore diameter fiber, resulting in an output energy of 16 µJper pulse. Assuming a tissue fluorescence efficiency inthe 400-nm region to be 0.01%, the average fluorescenceenergy that they calculated is ¾1.6 nJ. Given a collectionefficiency of 1% of the fluorescence escaping from thetissue surface to the detector fiber, the fluorescenceenergy coupled to the detector fiber is¾16 pJ. Accountingfor coupling losses and transmission efficiency through thespectrograph, the total energy pulse that reaches the PDAis two orders of magnitude lower, i.e. 400 fJ.

The signal-to-noise ratio obtainable by the PDAis determined by S, which is the total number ofphotoelectrons that reach the detector, and N, which isthe number of photoelectrons associated with noise. Thetotal noise generated by the various sources is definedaccording to Equation (6):

N D .SCN2R CN2

D/1/2 .6/

where the square root of S is the shot noise from the signalphotoelectrons collected by the detector, NR is the read-out noise inherent in the amplifier and electronics and ND

is the noise due to the dark current and is dependent onthe exposure time and temperature. Because the detectoris Peltier cooled to 5 °C during operation, the dark currentis negligible. The signal that reaches the photodetectoris estimated to be 400 fJ. Given a quantum efficiency inthe 400-nm region of 50% (1ð 1018 photoelectrons J�1),approximately 400 000 photoelectrons reach the PDA.The shot noise is calculated to be 632 electrons. The read-out noise for the PDA is 3000 electrons. Using thesequantities, the signal-to-noise ratio obtainable with theseparameters is approximately 130. Trujillo et al. have alsodone a systematic evaluation of the signal-to-noise ratioof a typical instrument used for fluorescence spectroscopyof tissue..71/ In addition, they have developed an analyt-ical expression to quantify the fluorescence efficiency ofthe tissue, which they have incorporated into the signal-to-noise ratio analysis of their fluorimeter.

4.6 Calculation of Tissue Fluorescence Efficiency

In order to calculate the expected signal-to-noise ratio ofa fluorimeter, two quantities are required: the throughputand inherent noise of the fluorimeter; and the fluorescenceefficiency of the tissue volume probed. The throughputand inherent noise of the fluorimeter is a function of thedelivery of the excitation light to the tissue surface, thecollection of the fluorescence from the tissue surface, andfinally conversion of the optical signal to an electronicsignal. The calculation of these three quantities hasbeen described in section 4.5. The fluorescence efficiencyis the quantitative relationship between the excitationand emission energy. Trujillo et al..71/ have developedan analytical expression to calculate the fluorescenceefficiency of tissue from fluorescence emission spectra oftissues measured in vivo.

The fluorescence efficiency (FE) of a turbid sample isdefined as the ratio of the total emitted photons to theexcitation photons and is described as shown in Equa-tion (7):

FE D∫ 1

0dz∫ 1lm

dlmHin.lx, z/∑

k

[µak.lx, z/

ð fk.lm, z/]Hout.lm, z/ .7/

where lx is the excitation wavelength, lm is the emissionwavelength, z is the depth within the sample, µa.lx, z/represents the absorption coefficient at the excitationwavelength of the fluorophore at depth z, f.lm, z/ is thefraction of the absorbed energy converted to fluorescenceat the emission wavelength of the fluorophore at depth z,and Hin.lx, z/ and Hout.lm, z/ represent the attenuationof the excitation and emission light, respectively, at depthz. The fluorescence efficiency is independent of the powerof the intensity at the excitation wavelength and the illu-mination/collection geometry. The fluorescence efficiencyof an unknown turbid sample can be calculated from flu-orescence measurements of the unknown sample andfluorescence measurements of an optically dilute homo-geneous standard with a known absorption coefficient(µa) and quantum efficiency using the same apparatus.Given the fluorescence emission spectrum of the turbidsample and that of a homogeneous standard with a knownµa and quantum efficiency, the fluorescence efficiency ofthe turbid sample can be expressed by Equation (8):

FEt.lx, lm/ D

∫ 1lm

Pt.lx, lm/ dlm∫ l

lm

Pstd.lx, lm/ dlm

ð∫ l

0dz∫ 1lm

dlmHin�std.lx, z/µa�std

ð.lx/fstd.lm/Hout�std.lm, z/�t

�std

.8/


where the subscript ‘‘t’’ corresponds to that of theturbid sample, the subscript ‘‘std’’ corresponds to thatof the homogeneous standard, Pt is the total integratedfluorescence power of the turbid sample, Pstd is thetotal integrated fluorescence power of the homogeneousstandard (both of which can be measured), and �t

and �std represent the collection efficiency of theturbid sample and homogeneous standard respectively.The first term in Equation (8) can be calculated fromthe ratio of the total integrated fluorescence powerof the turbid sample and the homogeneous standard.The second term requires knowledge of the opticalproperties as well as the collection geometry of thehomogeneous standard. The authors used a rhodaminesolution in a cuvette as the homogeneous standardin these calculations. Because the rhodamine standardused is an optically dilute homogeneous solution, andthe numerical aperture of the excitation and collectionis small (0.22), the attenuation of the excitation lightand fluorescent light in the rhodamine standard canbe described with the simple Beer–Lambert Law..2/

Hence, the wavelength-dependent µa can be calculatedsimply from the measured rhodamine absorbance. Thequantum efficiency of rhodamine is unity..72/ The doubleintegral in Equation (8) can be evaluated by using themeasured values of µa.lx/, µa.lm/ and f.lm/ of thehomogeneous standard. If f.lm/ is not known but thequantum efficiency is, then an alternative method canbe used to calculate the fluorescence efficiency, whichrequires changing f.lm/ into the quantum efficiency QE(defined by Equation 9):

QE D∫ 1lx

dlmf.lm/ .9/

Figure 13 displays the average fluorescence efficiencycalculated from the fluorescence emission spectra of atotal of 374 human cervical tissues at three excitationwavelengths in vivo. In Figure 13, the average fluores-cence efficiency values are plotted for normal squamoustissue, normal columnar tissues, low-grade squamousintra-epithelial lesions (SIL) and high-grade SIL atexcitation wavelengths 337, 380 and 460 nm. Standarddeviations are shown as error bars. For squamous cervi-cal tissues, the fluorescence efficiency decreases as tissueprogresses from a nondiseased normal squamous to a dis-eased state (low-grade and high-grade SIL). Furthermore,as the excitation wavelength increases, the fluorescenceefficiency decreases. The fluorescence efficiencies shownhere are instrument independent and can be used todetermine the signal-to-noise ratio needed to measurefluorescence emission spectra from tissues for differentinstrument configurations and excitation–emission col-lection geometries.

Ave

rage

FE

0NS LG HG NC NS LG HG NC NS LG HG NC

0.001

0.002

0.003

0.004

0.005

337 nm380 nm460 nm

Figure 13 Average fluorescence efficiency (FE) (šstandarddeviation) for human cervical normal squamous tissues (NS),normal columnar tissues (NC), low-grade SIL (LG) andhigh-grade SIL (HG) calculated from their fluorescence spectraat the excitation wavelengths 337, 380 and 460 nm. (Reproducedby permission of the Society for Applied Spectroscopy, fromMcCreery..70/)

5 CLINICAL APPLICATIONS

Currently, one of the most widely explored appli-cations of fluorescence spectroscopy is the detectionof endoscopically invisible, early neoplastic growth inepithelial tissue sites. Early neoplastic growth refers toprecancerous changes such as dysplasia and carcinoma insitu, which precede invasive cancer or carcinoma. Cur-rently, there are no effective diagnostic techniques forthese early tissue transformations. Fluorescence spec-troscopy is ideally suited for this application because ofits ability to examine tissue surfaces rather than tissuevolumes, and its adaptability to an endoscopic device.If fluorescence spectroscopy can be applied success-fully as a diagnostic technique in this clinical context,it may increase the potential for curative treatment andthus reduce complications. In addition to the potentialfor improved patient outcome, the fast and noninvasivenature of this diagnostic technique may also reduce healthcare costs.

Steady-state fluorescence measurements from small tis-sue regions (less than a few millimeters in diameter) aswell as steady-state fluorescence imaging of relativelylarge tissue fields (a few centimeters in diameter) havebeen performed..11,12/ To a much lesser extent, time-resolved fluorescence measurements are being exploredcurrently..11,12/ Furthermore, sources of both intrinsic(endogenous fluorophores) and extrinsic (exogenousfluorophores) fluorescence have been considered..11,12/

The advantage of using exogenous fluorophores is thatthe photophysical and pharmacokinetic properties can


be selected and are known. Furthermore, exogenousfluorophores are more highly fluorescent than endoge-nous fluorophores. On the other hand, the disadvantageof using exogenous fluorophores is that issues relatingto safety and toxicity of the drug being used have to beaddressed. Furthermore, the selection of the optimal timedelay after administration of the drug is nontrivial.

Several authors have reviewed the progress in thisfield over the last decade. Andersson-Engels et al..5,6/

discussed the possibilities of using fluorescence spec-troscopy for tissue diagnosis. The tissue types discussedwere malignant tumors and atherosclerotic lesions. Stud-ies of endogenous fluorescence as well as exogenousfluorescence with contrast agents were also presented. Ina subsequent review, Andersson-Engels et al..7/ reviewedthe scientific, technical and practical issues related to theuse of fluorescence spectroscopy for tissue diagnosis inclinical oncological applications. Papazoglou.8/ presenteda review on the diagnosis of malignancies and atheroscle-rotic plaques using fluorescence spectroscopy. Specialemphasis was given to problems that were raised duringclinical trials and recent experimental studies, such asthe identification of the fluorescent chemical species andthe determination of the illumination/collection geome-try for fluorescence spectroscopy of tissue. Bigio et al..9/

reviewed fluorescence spectroscopy and elastic scatter-ing spectroscopy of small regions of tissue, whereasAndersson-Engels et al..10/ focused on fluorescence spec-troscopy of relatively large tissue fields. Richards-Kortumet al..11/ and Wagnieres et al..12/ have written more com-prehensive reviews. The former review focuses on thequantitative aspects of fluorescence spectroscopy. Specif-ically, this review describes optical interactions pursuedfor biomedical applications, provides a descriptive frame-work for light interactions in tissue and, finally, reviewsthe important endogenous and exogenous molecules andhow they are exploited for tissue diagnosis. The latterreview, which focuses more on clinical aspects, presents acritical status report on the diagnosis of neoplastic tissuesin vivo using fluorescence spectroscopy.

5.1 Neoplasia

5.1.1 Endogenous Fluorescence

The phenomenon of fluorescence was first observed byStokes..73/ Much later, Stubel recognized the diagnosticpotential of tissue fluorescence..74/ Policard.47/ observedred fluorescence when examining necrotic tumors underillumination with UV and VIS light. Ronchese.48/ demon-strated in 1954 that ulceration is essential for theproduction of red fluorescence in human cutaneous,squamous cell carcinoma. The observed fluorescencewas attributed to endogenous porphyrins in the tis-sue. Ghadially et al..39,49 – 51/ also reported that ulcerated

squamous carcinomas exhibit a red fluorescence whenexposed to UV light. They concluded that this red flu-orescence may be due in part to the action of bacteriaon a protoporphyrin precursor. In 1965, Lycette et al..75/

suggested that fluorescence spectroscopy could be usedto discriminate between normal tissues and malignanttumors. Fluorescence emission spectra were recordedfrom excised normal tissue and malignant tumors of theesophagus, stomach, breast and thyroid at 330 nm exci-tation. All tissues fluoresced in the range 360–600 nm. Itwas found that the fluorescence intensities of malignanttumors were less than that of normal tissue from thesame patient. Subsequently, the groups of Profio,.76,77/

Alfano,.78/ Lohmann.79/ and Yang.80/ did pioneering stud-ies on in vitro and in vivo fluorescence spectroscopy ofneoplastic and nonneoplastic animal and human tissues.

Fluorescence spectroscopy has been used to evaluateneoplastic and nonneoplastic tissues in vitro and in vivo.Generally, the results of in vitro studies have been usedto guide the design of experimental parameters for invivo studies..81/ However, in vitro studies can introducesignificant artifacts due to hemoglobin reabsorption andoxidation of certain fluorophores,.82/ questioning therelevance of in vitro studies to in vivo studies. Richards-Kortum et al..11/ review in vitro and in vivo studies onfluorescence spectroscopy of neoplasia and in vitro studieson fluorescence spectroscopy of atherosclerosis. Thisreview covers only in vivo studies, in order to demonstratethe diagnostic potential of fluorescence spectroscopy inclinical applications.

UV/VIS fluorescence spectroscopy can be developedand employed to differentiate diseased from nondiseasedtissues in vivo. The altered biochemical and morpho-logical state that occurs as tissue progresses from anondiseased to a diseased state is reflected in the spectralcharacteristics of the measured fluorescence. This spec-tral information can be compared to tissue histology – thecurrent gold standard – which indicates the absence orpresence and grade of disease. Mathematical algorithmsthen can be developed and optimized to classify tissuesinto their respective histological category based on theirspectral features. These mathematical algorithms can beimplemented in software, thereby potentially enablingfast, noninvasive, automated screening and diagnosis ina clinical setting. Although the signal-to-noise ratio offluorescence spectroscopy is several orders of magnitudelower than that associated with absorption spectroscopyover the same wavelength range, this technique affordshigher sensitivity and specificity for biochemical charac-terization.

Table 2.83 – 100/ summarizes the key aspects of a rep-resentative set of clinical applications of fluorescencespectroscopy (the most recent investigation per group andinvestigations from groups who reported the performance


Table 2 Fluorescence spectroscopy of neoplastic and nonneoplastic tissues in vivo obtained from a representative set of clinicalapplicationsa

Tissue Excitation Measurement Dimension reduction: Classification: Sample size SE; SPwavelength (nm) type emission wavelengths (nm) decision scheme ND; D (%)

Colon.83/ 337 Spectra MVLR (spectra) Binary 121; 56 80; 92Colon.84/ 337 Spectra MVLR (spectra) Binary 182; 32 100; 84Colon.85/ 370 Spectra I (460, 600, 680) Bayes theorem 67; 32 94; 91b

Colon.86/ 360 Images I (>400) Binary 12; N/A 83; N/AColon.87/ 337 Transient decay t (550) Binary 13; 11 85; 87Cervix.88/ 337, 380, 460 Spectra PCA (spectra) Bayes theorem 122; 59 82; 68b

Cervix.88,89/ 337, 380, 460 Spectra PCA of 15 I.lx,lm/ Bayes theorem 122; 59 84; 65b

15 I.lx,lm/ Neural network 91; 67b

Bronchus.90/ 442 Images I (red, green) Binary 558; 142 67; N/Ac

Bladder.91/ 337 Spectra I (385, 455) Binary 85; 29 97; 98d

Bladder.92/ 308 Spectra I (360, 440) Binary 35; 31 100; 100d

Esophagus.93/ 410 Spectra I (480, 680) Binary 252; 56 40; 96Oral cavity.94/ 370 Spectra I (490) Binary 100; 87.5

410 I (640)Oral cavity.95/ 337 Spectra I (peak) Binary 17; 11 94; 100

410 I (635, 490)Head and neck.96/ 442 Images I (red, green) Binary 16; 16 100; 87.5d

Head and neck.97/ 375–440 Images, spectra I (>515) Binary 36; N/A 43.3; N/Ad

36; N/A 94.4; N/Ad

Larynx.98/ 442 Images I (red, green) Binary 328 total 72.5; 94c,d

Larynx.99/ 442 Images I (red, green) Binary 87; 28 85; 87c,d

Skin.100/ 336 Images I (475) Binary 318; 90d 83; 79

a Acronyms: I D intensity; t D decay time; MVLR D multivariate linear regression; PCA D principal component analysis; lx D excitationwavelength; lm D emission wavelength; ND D nondiseased; D D diseased; SE D sensitivity; SP D specificity; N/A D not applicable.

b Sensitivity and specificity were evaluated prospectively.c Sensitivity and specificity were evaluated for a combination of fluorescence and conventional endoscopic techniques.d Sensitivity and specificity were calculated for the discrimination between cancers and noncancers (other investigations calculated the sensitivity

and specificity for discriminating precancers and cancers from normal tissues).

of their technique). In these applications, fluorescencespectroscopy in the UV/VIS spectral regions was usedfor the detection of neoplastic tissues in vivo. In par-ticular, Table 2 provides the organ sites studied, theexcitation wavelength(s) used, the type of measurementsthat were made, the dimensionally reduced spectral vari-ables and the corresponding classification scheme usedin the mathematical algorithm, the sample size for thediseased and nondiseased populations and the corre-sponding sensitivity and specificity. Note that in all theseclinical applications only the endogenous fluorescence,absorption and scattering properties of the tissue wereexploited.

The neoplastic tissues that were evaluated spec-troscopically are from the colon,.83 – 87/ cervix,.88,89/

bronchus,.90/ bladder,.91,92/ brain.101/ (preliminary studiesonly), esophagus,.93/ oral cavity,.94,95/ head and neck,.96,97/

larynx,.98,99/ skin,.100/ bile duct.102/ (preliminary studiesonly), stomach.103/ (in vitro studies only) and breasttissues.104 – 107/ (in vitro studies only). The excitation wave-lengths that were selected correspond to those used toexcite fluorophores in the UV/VIS spectral regions (seeTable 1). Most groups measured fluorescence emission

spectra from tissue sites that are 1–2 mm in diameter.Only a few groups measured fluorescence images fromtissue regions that are a few centimeters in diameter,and only one group measured the transient fluorescencedecay from colon tissues in vivo.

There are generally two steps involved in the devel-opment of a mathematical algorithm that is based onfluorescence spectroscopy. The first part is to reducedimensionally the measured spectral variables, and thesecond step is to develop a classification scheme for thediscrimination of these useful spectral parameters intorelevant histological/histopathological categories. Thedevelopment of current mathematical algorithms basedon fluorescence spectroscopy can be classified broadlyinto three categories: algorithms based on qualitativelyselected spectral variables (fluorescence intensities atseveral emission wavelengths); algorithms based on sta-tistically selected spectral parameters (a more robustuse of all the measured spectral information); and algo-rithms based on parameters that reflect the biochemicaland/or morphological features of the tissue. Classificationschemes have been primarily based on a binary dis-crimination or a probability-based classification scheme.


Mostly, algorithms have been based on qualitatively orstatistically selected spectral variables with binary classi-fication schemes as indicated in Table 2.

The sensitivity is defined as the fraction of dis-eased samples correctly classified, and the specificityis defined as the fraction of nondiseased tissues cor-rectly classified. In several clinical studies, the sensitivityand specificity were evaluated prospectively rather thanretrospectively to obtain an unbiased estimate of theperformance of this technique..85,88,89/ In some cases thesensitivity and specificity were evaluated for a combi-nation of fluorescence spectroscopy and conventionalendoscopy..90,98,99/ Finally, in the case of the bladder,head and neck and larynx, the sensitivity and specificityare reported for the discrimination between cancers andnoncancers..91,92,96 – 99/ In other clinical applications, thesensitivity and specificity are reported for the discrimi-nation of precancers and cancers from normal tissues. Inthe majority of clinical studies performed the sensitivityand specificity are greater than 80%, reflecting the highclassification accuracy of fluorescence spectroscopy forthe detection of neoplastic tissues in vivo. The sensitivityand specificity reported here are similar or superior tocurrent clinical modalities that are used routinely.

5.1.2 Exogenous Fluorescence

There are a number of groups that have exploredthe diagnostic potential of fluorescence spectroscopyof exogenous fluorophores (particularly photosensitizingagents developed for PDT) in tissues in vivo..12/ Auleret al..108/ and Figgie.109/ first observed red fluorescencefrom animal tumors after administration of exogenousporphyrins. The first use of fluorescin to improve thedetection and identification of brain tumors in vivowas reported by Moore et al..110/ The accumulation ofhematoporphyrin in various types of cancers was discov-ered and exploited during the 1950s..111 – 113/ HpDs wereevaluated subsequently for localization in cancers of var-ious organs, including the esophagus, tracheobronchialtree and cervix,.114 – 120/ and for characterizing suspicioushead and neck lesions..121/ The first cystoscopic obser-vations of HpD fluorescence in urinary bladder tumorswere reported by Kelly et al..122/ These and other initialclinical and animal model studies of HpD-mediated PDTled to fluorescence diagnostic applications by many othergroups using this exogenous fluorophore. Subsequently,Kreigmair et al..123/ proposed the use of 5-ALA-inducedPpIX fluorescence for the detection of bladder neoplasms.

Photosensitizing agents, particularly HpD and 5-ALA,have been used as contrast agents for fluorescencespectroscopy of neoplastic tissues in a wide variety oftissue sites, including the skin, bladder, bronchus, colon,esophagus, head and neck, and breast..12/ However,

because most of these exogenous fluorophores rely onthe differences in the vasculature between diseased tissues(which generally exhibit leaky vessels) relative to theirnondiseased counterparts, it remains to be seen whetherthe characteristics that determine localization of theexogenous fluorophore in cancers are also present inprecancerous tissues.

5.2 Atherosclerosis

Edholm et al..124/ suggested that quantitative opticaltechniques could be used to improve the detection ofatherosclerosis in vivo; they measured the reflectance ofthe aorta at 500 and 550 nm in 15 patients undergoingaortography. Kitrell et al..125/ demonstrated that fluores-cence spectroscopy could be used to discriminate betweennormal aorta and fibrous plaque, and suggested that opti-cal diagnosis (using low-power illumination) and therapy(using high-power illumination) of atherosclerosis couldbe combined in a single fiber-optic device. Subsequently,a number of groups investigated the utility of fluorescencespectroscopy for the diagnosis of atherosclerotic plaque,many with the goal of developing a guidance system forlaser angiosurgery catheters..8,11/

The fluorescence emission spectra of normal artery tis-sue and atherosclerotic plaques have been measured overthe entire range of UV/VIS excitation wavelengths..11/

However, most of these investigational studies to demon-strate the diagnostic potential of fluorescence spec-troscopy have been performed in vitro on excised tissuesand hence have not addressed the problems of mak-ing these measurements in vivo on the intact arterywall.

6 CLINICAL INSTRUMENTS

In this section representative examples of instrumentsused in clinical applications of fluorescence spectroscopyare described. In particular, details of the instruments andtypical data are presented to provide the reader with anunderstanding of the main characteristics of instrumentsfor UV/VIS fluorescence spectroscopy.

6.1 Single-pixel, Three-excitation-wavelengthFluorimeter

Although fluorescence emission spectra of normal tissue,dysplasia and invasive carcinoma have been measuredpreviously from several organ sites in vivo at single exci-tation wavelengths (see Table 2), Ramanujam et al..88/

were among the first to address the necessity to mea-sure fluorescence emission spectra at several excitation


wavelengths in order to exploit more fully the tis-sue biochemistry and morphology. They developed aportable fluorimeter that can be used to measure flu-orescence emission spectra from 1-mm-diameter humancervical tissue sites in vivo at three excitation wavelengthsin the UV/VIS spectral regions. These wavelengthswere selected primarily because they coincide with theabsorption bands of biologically relevant fluorophores,namely the metabolic coenzymes and structural proteinsand chromophores, which include oxygenated and deoxy-genated hemoglobin. A schematic of the fluorimeter isshown in Figure 14.

Two nitrogen pumped-dye lasers (5 ns pulse duration,30 Hz repetition rate) are used to provide illuminationat three different excitation wavelengths: one lasergenerates light at 337 nm (fundamental) and has a dyemodule in a resonant cavity that can be used to generatelight at 380 nm using the fluorescent dye BBQ. The dyemodule in the resonant cavity of the second laser isused to generate light at 460 nm, using the fluorescentdye Coumarin 460. Laser illumination at each excitationwavelength of 337, 380 and 460 nm is coupled to eachof three excitation fibers in a fiber-optic probe. Notethat two 10-nm bandpass filters – one centered at 380 nmand the other centered at 460 nm – are placed betweenthe excitation fiber and the two dye modules, to preventleakage from the fundamental at 337 nm.

The fiber-optic probe consists of a central fibersurrounded by a circular array of six fibers; all seven fibershave the same characteristics (0.22 numerical aperture,200 µm core diameter, 245 µm diameter with cladding).Three fibers along the diameter of the distal end of theprobe are used for excitation light delivery. The purpose

of the remaining four fibers is to collect the emitted lightfrom the tissue area directly illuminated by the excitationlight..57,58/ A quartz shield (3 mm in diameter and 2 mmthick) at the tip of the distal end of the probe, which isin direct tissue contact, provides a fixed distance betweenthe fibers and the tissue surface so that the fluorescenceintensities can be measured in calibrated units. A tissuearea that is 1 mm in diameter is illuminated by eachexcitation delivery fiber. The average energies per pulseon the tissue surface at 337, 380 and 460 nm excitationwere 15.2, 11.5 and 18 µJ mm�2, respectively, in this case.

The proximal ends of the four emission collectionfibers are arranged in a circular array and imaged atthe 500-µm-wide entrance slit of an f 3.8 spectrographequipped with a 300 grooves mm�1 grating coupled toa 1024-intensified PDA controlled by a multichannelanalyzer. The collection optics between the proximalend of the four fibers and the spectrograph are two,fused-silica, planoconvex lenses. Between these lenses isa filter wheel assembly containing long-pass filters with50% transmission at 360, 400 and 475 nm, which areused to reject backscattered excitation light at 337, 380and 460 nm excitation, respectively, from the detector.The nitrogen pumped-dye lasers are used for externaltriggering of a pulser, which serves to synchronize the200-ns collection gate of the detector to the leadingedge of the laser pulse. The use of gated detection inconjunction with pulsed excitation eliminates the effectsof ambient light during fluorescence measurements. Thetotal time required to record fluorescence emissionspectra at all three excitation wavelengths from one tissuesite was approximately 5 s. Spectra were collected in theVIS spectral region, with a resolution of 10 nm (fwhm)

Intensifieddiode array

Gate pulser Controller Computer

Spectral resolution: 10 nm

Polychromator

30 Hz rep rate5 ns pulse duration

Nitrogenpumped dye

laserNitrogen

pumped dyelaser

337 nm

380 nm

460 nm

Collection fibers

Excitation fibers

Probe

Excitation fibers

Collection fibers Quartzshield

(b) (c)

(a)

Figure 14 Schematic of a three-excitation-wavelength system. (Reproduced by permission of the American Society forPhotobiology, from Ramanujam et al..88/)


and a signal-to-noise ratio of 100 : 1 at the fluorescencemaximum at each excitation wavelength. Data acquisitionand storage are achieved using a computer.

All spectra were corrected for the nonuniform spectralresponse of the detection system using correction fac-tors obtained by recording the spectrum of a NationalInstitute of Standards and Technology (NIST) traceable,calibrated, tungsten ribbon filament lamp. Spectra fromeach cervical tissue site at each excitation wavelengthwere averaged to obtain a single spectrum per site.The average tissue spectra were then normalized to theaverage peak fluorescence intensity of the rhodamine610 calibration standard at the corresponding excita-tion wavelength for that patient; absolute fluorescenceintensities are reported in these calibrated units.

Figure 15 illustrates average fluorescence emissionspectra per site acquired from cervical tissue sites at337, 380 and 460 nm excitation from a typical patient.All fluorescence intensities are reported in the sameset of calibrated units. Evaluation of the tissue spectraat 337 nm excitation indicates that the fluorescence

3500.0

0.1

0.2

0.3

0.4

0.5

400 450 500 550 600

Spectra at 337 nm excitation

NCHG

LG

NS

(a)

4000.0

0.1

0.2

0.3

450 500 550 600 650


NS

NC

HG

HG

LG

Flu

ores

cenc

e in

tens

ity

(b)

(c)

0.3

0.2

0.1

0.0460 510 560 610 660

Wavelength (nm)


NS

LGNC

Figure 15 Fluorescence spectra measured at 337, 380 and460 nm excitation from the cervix of a typical patient usingthe three-excitation-wavelength fluorimeter. (Reproduced bypermission of the American Society for Photobiology, fromRamanujam et al..88/)

intensity of the precancerous SIL is less than that ofthe corresponding normal squamous tissue and greaterthan that of the corresponding normal columnar tissue.Evaluation of the spectra at 380 nm excitation indicatesthat the fluorescence intensity of SIL is less than that ofthe corresponding normal squamous tissue, with the low-grade SIL exhibiting the weakest fluorescence intensity.Note that the fluorescence intensity of the normalcolumnar tissue is indistinguishable from that of thehigh-grade SIL. Finally, evaluation of spectra at 460 nmexcitation indicates that the fluorescence intensity of SILis less than that of the corresponding normal squamoustissue and greater than that of the corresponding normalcolumnar tissue. The spectra shown here demonstratethat it is possible to obtain fluorescence emission spectrafrom tissues in vivo at several excitation wavelengths, witha high signal-to-noise ratio and in a fast and noninvasivemanner.

6.2 Single-pixel, Excitation–Emission Matrix System

It is apparent that fluorescence emission spectra atmultiple excitation wavelengths spanning the UV/VISspectral regions are needed to characterize properlythe patho-physiologically relevant, endogenous fluo-rophores in tissue. Furthermore, reflectance measure-ments at these wavelengths are also needed to probetissue absorption and scattering properties and enabledeconvolution between the fluorescence, absorption andscattering..126,127/ Zeng et al..128/ developed an instrumentto measure the fluorescence emission spectra and diffusereflectance spectra from the human skin, simultaneously.However, this instrument was capable of measuring flu-orescence emission spectra at only a single excitationwavelength of 380 nm.

Zangaro et al..69/ integrated and improved upon thedesigns of Ramanujam et al..88/ and Zeng et al..128/ byconstructing a fast EEM system that can measure fluores-cence emission spectra at multiple excitation wavelengthsand diffuse reflectance spectra from 1- to 2-mm-diametertissue sites in vivo quickly and noninvasively. Prelimi-nary considerations for the design of this system wereas follows: rapid wavelength tunability and real-timedata acquisition for fluorescence spectroscopy at multi-ple excitation wavelengths; the ability to measure diffusereflectance in addition to fluorescence; gated detectionof spectral measurement with minimal interference fromambient light; and capability for optical fiber deliveryand collection. The system, which was designed withthe aforementioned considerations in mind, is shown inFigure 16.

A nitrogen pumped-dye laser is used as the excitationlight source. Delivery of light to tissue and collectionof returned spectral information is accomplished by


Detectorcontroller Filter wheel

Spectro-graph

Pulsegenerator

Nitrogenlaser

Dyewheel

Whitelight

AC motorL1

M1

L4

L2 L3Prism

M2

Fiber-optic/catheter

Trig

Trig

Detector

Microcomputer

Sample

TrigSync

Figure 16 EEM system designed to measure tissue fluores-cence spectra at multiple excitation wavelengths and spectrallyresolved broad band diffuse reflectance spectra in vivo. (Repro-duced by permission of the Optical Society of America, fromZangaro et al..69/)

means of an optical fiber probe. A filter wheel placed infront of the spectrograph rejects backscattered excitationlight. The fluorescence is dispersed in the spectrograph,detected by an intensified PDA, coupled to an opticalmultichannel analyzer and stored in a computer. Thenovel aspects of this system, which lends itself tomulti-excitation fluorescence spectroscopy and diffusereflectance spectroscopy of tissues in the UV/VIS spectralregions, are described below.

Dye cuvettes (with appropriate dyes) mounted on arotating wheel that rapidly traverses a single resonantcavity are capable of generating multiple wavelengths.Also, a corresponding filter wheel placed in front of thespectrograph allows rejection of backscattered excitationlight at each wavelength when synchronized with the dyewheel. Furthermore, a 10-µs pulsed xenon lamp is alsoincorporated such that diffuse reflectance measurementscan be made over the UV/VIS spectral regions. Fiber-optic delivery and collection have been designed toprovide for the remote operation of the system and forthe measurement of fluorescence and diffuse reflectancedirectly from the tissue area that is illuminated, inorder to remove the effects of the illumination/collectiongeometry..57,58/ Furthermore, the fiber-optic probe wasdesigned such that the diffuse reflectance and fluorescencecan be made from the same tissue site. The acquisitiontime for both fluorescence and reflectance measurementsis less than 1 s.

The performance of this system has been validatedclinically by passing the fiber-optic probe through oneof the accessory channels of a colonoscope and bringing

400 500 600 700

400 500 600 700

400 500 600 700

400 500 600 700

400 500 600 700 400 500 600 700

400 500 600 700

400 500 600 700

400 500 600 700

400 500 600 700

43210

2

1

0

2.01.51.00.50.0

3

2

1

0

1.5

1.0

0.5

0.0

3

2

1

0

1.5

1.0

0.50.0

1.5

1.0

0.5

0.0

1.0

0.50.0

2.01.51.0

0.50.0

exc = 337 nm exc = 360 nm




exc = 485 nmexc = 506 nm

Wavelength (nm) Wavelength (nm)(a)

Nor

mal

ized

fluo

resc

ence

inte

nsity

2.01.5

0400 500 600 700

0.05

0.10

Diff

use

refle

ctan

ce

Wavelength (nm)(b)

Figure 17 (a) Fluorescence spectra at 10 excitation wave-lengths and (b) diffuse reflectance spectrum over the UV/VISspectral regions measured from a colon tissue site in vivo.(Reproduced by permission of the Optical Society of America,from Zangaro et al..69/)

it into flush contact with the tissue surface. Figure 17(a)displays fluorescence emission spectra at multiple exci-tation wavelengths and Figure 17(b) displays a diffusereflectance spectrum from colon tissue. Figure 17(a)shows spectra measured at 10 excitation wavelengths in


the UV/VIS spectral regions. The spectra at the differentexcitation wavelengths appear to correspond with spectraof biologically relevant fluorophores, including collagen(337 nm excitation), reduced PN (360 nm excitation) andoxidized Fp (428 nm excitation). Note that the effectsof hemoglobin absorption on the fluorescence emissionspectra are observed at 337 nm excitation. Figure 17(b)shows the diffuse reflectance spectrum obtained from thesame tissue site. The spectrum exhibits the absorptionbands of hemoglobin at 420 nm (Soret band) and at 540and 580 nm (a and b bands).

Zuluaga et al..129/ have also developed a fast EEMsystem that measures fluorescence emission spectra atmultiple excitation wavelengths and diffuse reflectancespectra rapidly (4 min total measurement time) and non-invasively from tissues in vivo. Specifically, the systemmeasures the fluorescence EEM at 330–500 nm excita-tion and 380–700 nm emission. The diffuse reflectancespectra are measured at 380–950 nm. The primary differ-ence between this system and that developed by Zangaroet al..69/ is that it employs a white light source with a seriesof bandpass filters instead of a nitrogen pumped-dye laserfor generating multiple excitation wavelengths for fluo-rescence spectroscopy. Furthermore, Zuluaga et al..129/

present a method based on autocorrelation vectors toidentify the excitation and emission wavelengths wherethe spectra of diseased and nondiseased tissues differ themost.

6.3 Fluorescence Imaging Systems

Several groups have developed endoscopic-compa-tible.86,97,130 – 134/ and nonendoscopic-based.100,135,136/ flu-orescence imaging systems. The most notable is thatoriginally developed by Palcic et al. for fluorescencebronchoscopy,.130/ which has now led to a commerciallight-induced fluorescence endoscopy (LIFE) device (Xil-lix Technologies Corporation, Richmond, BC, Canada)that is used for fluorescence imaging of relatively largetissue fields (a few centimeters in diameter). The LIFEdevice consists of a white light source and a color CCD forthe acquisition of white light images, a helium–cadmiumlaser and two filtered, high-sensitivity CCD cameras forthe acquisition of green and red fluorescence images, anendoscope-compatible fiber-optic bundle and a computerwith a monitor. The laser is used to provide excitationlight at 442 nm. The fluorescence emitted is collected andcoupled via an imaging, fiber-optic bundle to two fil-tered, image-intensified CCD cameras (one for the greenwavelength band at 480–520 nm and one for the redwavelength band at 630 nm and longer). The fluorescenceimage is digitized using an imaging board in the com-puter. Using a mathematical transformation (a nonlineardiscriminant function combination of the red and green

intensity images), a pseudo-image of the observed tissuesite is obtained and displayed on the monitor in real-time. Normal tissues appear green and neoplastic tissuesappear brownish-red.

The LIFE device has been tested in a multicenterclinical trial to evaluate if fluorescence bronchoscopy,when used as an adjunct to white light bronchoscopy, canimprove the bronchoscopist’s ability to locate areas suspi-cious of dysplasia for biopsy and histological examinationas compared to white light bronchoscopy alone..90/ TheLIFE device also has been used successfully to evaluateneoplastic lesions in other organ sites, including the headand neck,.96/ larynx,.98,99/ bile duct,.102/ and gastrointesti-nal tract..137,138/

Recently, Zeng et al..134/ elaborated on the fluores-cence imaging system for the gastrointestinal tract. Thedevice consists of a mercury arc lamp, two intensifiedCCD cameras, a fiber-optic endoscope and a computer-based console. The system is capable of working in threedifferent modalities: conventional white light imagingmode; light-induced fluorescence imaging mode, based onfluorescence imaging of two wavelength bands in the redand green; and light-induced fluorescence and reflectanceimaging mode, based on the combination of red and greenfluorescence images and a red-NIR reflectance image.

7 METHODS OF ANALYSIS

Most screening and diagnostic algorithms developed fromfluorescence spectroscopy of tissues incorporate qualita-tively or statistically selected spectral variables, whichare evaluated using a binary or probability-based classi-fication scheme. For example, Panjehpour et al..93/ havedeveloped an algorithm that uses qualitatively selected,fluorescence intensities at several emission wavelengthsin a binary classification scheme for the detection of Bar-rett’s esophagus in vivo. On the other hand, Ramanujamet al..88/ have developed an algorithm that uses statis-tically selected spectral variables and probability-basedclassification for cervical precancer detection in vivo.This multivariate statistical algorithm employs PCA.139/

to reduce dimensionally the preprocessed tissue fluores-cence emission spectra into orthogonal principal compo-nents with minimal information loss. Bayes theorem.140/

is then used to develop probability-based classificationusing the diagnostically relevant principal components.The advantage of using statistical rather than qualitativeanalysis of the tissue fluorescence emission spectra is thatthe entire spectral information content can be exploited.Furthermore, the benefit of using a probability, ratherthan a binary, classification scheme is that the likelihoodof the classification being correct is also provided.


An example of an algorithm that uses biochemicaland/or morphological features that are related to thetissue fluorescence emission spectra, coupled with aprobability-based classification, is that developed byRichards-Kortum et al..141/ This algorithm discriminatesbetween normal coronary arteries and noncalcified andcalcified atherosclerotic plaque in vitro. The contributionof the biochemical and/or morphological features isextracted from the tissue fluorescence emission spectravia an analytical model of tissue fluorescence, based onthe exponential attenuation of light in an absorption-dominant medium. Bayes’ theorem is then used forprobability classification.

The advantage of using a physically based model overa statistically based model is that the former methodprovides insight into the biochemical and morphologicalbasis. However, there are notable advantages to usinga statistical model such as PCA. Although it is alinear method of analysis, PCA can still be used tomodel effectively the nonlinear turbid tissue fluorescence.Because it is less restrictive, PCA can permit a betterfit to the fluorescence emission spectra than simplephysical models can, through the use of a linearcombination of orthogonal principal components..142/ Inaddition, the principal components are not correlatedwith each other and therefore can be used in avariety of classification algorithms that generally requireun-correlated variables. Furthermore, each principalcomponent can be correlated to the spectral variablesof the original tissue fluorescence emission spectra, thusproviding insight into the spectral features that contributeto the classification.

Most algorithms to date use qualitatively and, to alesser extent, statistically selected variables with binaryor probability-based classification schemes as indicatedin Table 2. The development of a physically based modelthat uses biochemical and morphological features thatare related to the measured tissue fluorescence emissionspectra has been hampered by the fact that fluorescencespectroscopy of human tissue is greatly affected by theabsorption and scattering of the excitation light and theemitted light, making interpretation of the measuredspectral information challenging..53/

Given the high classification accuracy that can beachieved using statistically based algorithms,.88/ and thedifficulties associated with the development of physicallybased algorithms,.141/ it is perhaps worthwhile consideringthe development of a hybrid algorithm that incorporatesthe key features of both. For example, the statisticalmodel could be related to the physical model in order torealize the biochemical and/or morphological basis of thestatistically selected spectral variables that are used forclassification purposes.

Section 7.1 provides a detailed description of thealgorithm developed by Ramanujam et al.,.88/ which usesstatistically selected spectral parameters in conjunctionwith a probability based classification process. Section 7.2discusses physically based models to analyze tissuefluorescence emission spectra and their potential toelucidate the biochemical and morphological featuresthat contribute to the tissue fluorescence emission spectra.Section 7.3 presents various approaches to account for theillumination/collection geometry-related distortion of thetissue fluorescence emission spectra measured in vivo.

7.1 Statistically Based Model

Ramanujam et al..88/ developed a formal analytical pro-cess for the development of screening and diagnosticalgorithms for the detection of human cervical precanceror SIL in vivo. The formal analytical process is displayed inFigure 18, where the text in the dashed-line boxes repre-sents the mathematical steps implemented on the spectraldata, and the text in the solid-line boxes represents theoutput after each mathematical process. There are fourprimary steps involved in the multivariate statistical anal-ysis of tissue fluorescence emission spectra. The firststep is to preprocess spectral data to reduce interpatientand intrapatient variation within a tissue type; the pre-processed spectra are then dimensionally reduced to aninformative set of principal components that describemost of the variance of the original spectral data set usingPCA..139/ Next, the principal components that containdiagnostically relevant information are selected usingan unpaired, one-sided Student’s t-test, and finally aclassification algorithm based on Bayes theorem.140/ isdeveloped using these diagnostically relevant principalcomponents.

In summary, three constituent algorithms were devel-oped using multivariate statistical analysis: constituentalgorithm 1 discriminates between SIL and normalsquamous tissues; constituent algorithm 2 discriminatesbetween SIL and normal columnar tissues; and algo-rithm 3 differentiates high-grade SIL from low-grade SIL.The three constituent algorithms were then combinedto develop two composite algorithms: constituent algo-rithms 1 and 2 were combined to develop a compositescreening algorithm that discriminates between SIL andnon-SIL. All three constituent algorithms were then com-bined to develop a composite diagnostic algorithm thatdifferentiates high-grade SIL from non-high grade SIL.

Inputs into the multivariate statistical algorithmincluded the preprocessed fluorescence spectra at all threeexcitation wavelengths (full-parameter) and fluorescenceintensities at a reduced number of excitation–emissionwavelength pairs (15 reduced parameters) selected fromthe component loadings calculated from PCA..139/ The


PREPROCESSING

Normalized spectra atthree excitation wavelengths

Normalized, mean-scaled spectraat three excitation wavelengths

DIMENSION REDUCTION: PRINCIPAL COMPONENT ANALYSIS

SELECTION OF DIAGNOSTIC PRINCIPAL COMPONENTS: T-TEST

CLASSIFICATION: LOGISTIC DISCRIMINATION

DEVELOPMENT OF COMPOSITE ALGORITHMS

Constituent Algorithm 3 Constituent Algorithm 2Constituent Algorithm 1

Posterior probabilityof being NS or SIL

Posterior probabilityof being LG or HG

Posterior probabilityof being NC or SIL

Posterior probability of beingSIL or non-SIL

Posterior probability of beingHG SIL or non-HG SIL

(1, 2, 3) Composite diagnosticalgorithmComposite screening algorithm (1, 2)

Figure 18 Formal analytical process for the development of screening and diagnostic algorithms for the differential detection ofcervical pre-cancer or SILs. NS D normal squamous; NC D normal columnar; LG D low grade; HG D high grade. (Reproduced bypermission of the American Society for Photobiology, from Ramanujam et al..88/)

Table 3 Comparison of accuracy of composite screening and diagnostic (full and reduced-parameter) algorithms to that of Papsmear screening and colposcopy in expert hands

Classification SIL v. Non-SIL HG SIL v. Non-HG SIL

Sensitivity (%) Specificity (%) Sensitivity (%) Specificity (%)

Pap smear screening 62š 23 68š 21 N/A N/AColposcopy in expert hands 94š 6 48š 23 79š 23 76š 13Full-parameter composite algorithm 82š 1.4 68š 0.0 79š 2 78š 6Reduced-parameter composite algorithm 84š 1.5 65š 2 78š 0.7 74š 2

HG D high grade; N/A D not applicable.

algorithm was developed on a calibration set and testedon a prediction set of approximately equal numbers ofsamples in each tissue category.

Table 3 compares the sensitivity and specificity of thecomposite screening and diagnostic algorithms with thatof Pap smear screening.143/ and colposcopy in experthands..144/ Table 3 indicates that the composite screen-ing algorithm has a similar specificity and a significantlyimproved sensitivity relative to Pap smear screening. Acomparison of the composite screening algorithm to thatof colposcopy in expert hands for differentiating SILfrom non-SIL indicates that this algorithm demonstratesa 10% decrease in sensitivity but a 20% improvementin specificity. The composite diagnostic algorithm andcolposcopy in expert hands both discriminate high-gradeSIL from non-high-grade SIL with a similar sensitivityand specificity. Finally, the sensitivity and specificity ofthe reduced-parameter algorithms that use an order ofmagnitude fewer variables is within 5% of the sen-sitivity and specificity reported for the full-parameter

algorithms. This has important implications in using fluo-rescence spectroscopy to identify fluorescence intensitiesat a reduced number of optimal excitation–emissionwavelength pairs for the implementation of fluorescenceimaging.

7.2 Physically Based Models

Although it has been demonstrated that fluorescencespectroscopy can be used to differentiate diseased fromnondiseased tissues with high sensitivity and specificity,the underlying biochemical and morphological basis forthe spectral differences is poorly understood. This hasbeen hampered by the fact that fluorescence spectroscopyof a turbid medium such as tissue is complicated by itsabsorption and scattering properties..53/

A fluorescence emission spectrum measured fromtissue over the UV/VIS spectral region is primarilyattributed to the superposition of the fluorescence ofa variety of biological molecules that contain naturallyoccurring fluorophores. Fluorescence emission spectra


measured from tissues also contain information aboutthe absorbing and scattering properties of that medium.Quantification of the concentration and distributionof fluorophores within tissue in principle involves:deconvoluting the absorption and scattering propertiesof the tissue from the measured fluorescence emissionspectrum; and quantifying the identity and distributionof the fluorophores that contribute to the deconvolvedspectrum. The former step requires the developmentof a transfer function based on the measurementsof tissue optical properties, whereas the latter steprequires knowledge of the identity and distribution ofthe fluorophores within the tissue.

7.2.1 Transfer Function

Deconvolution of the absorption and scattering prop-erties of the tissue from the measured fluorescenceemission spectrum requires knowledge of the tissueoptical properties, i.e. the absorption coefficient, thescattering coefficient and the anisotropy parameter. Ana-lytical methods based on Kubelka–Munk theory.61/ andthe adding–doubling method.62/ have been developed tocalculate tissue optical properties in the UV/VIS spectralrange. However, these models require the measurementof diffuse transmittance and reflectance from tissues. Dif-fuse transmittance measurements cannot be made fromtissues in vivo, thus limiting the number of parametersneeded to determine tissue optical properties using thesesimple analytical techniques.

Several groups have developed simple analytical mod-els based on measurements of fluorescence and diffusereflectance only (no transmittance measurements areneeded) to deconvolve absorption and scattering proper-ties of the tissue from the measured fluorescence emissionspectrum..126,127/ These analytical techniques, althoughsimplistic, do provide a first step towards quantifying thebiochemical and morphological characteristics of tissuefluorescence emission spectra.

In particular, Gardner et al..127/ have developed ananalytical expression for recovering the intrinsic fluo-rescence coefficient (which is defined as the product ofthe fluorophore absorption coefficient and the fluores-cence quantum yield) of a homogeneous turbid mediumfrom a surface measurement of fluorescence and dif-fuse reflectance. The intrinsic fluorescence coefficientb.lx, lm/ is defined simply by Equation (10):

b.lx, lm/ D µa.lx/f.lm/ .10/

where lx is the excitation wavelength, lm is the emissionwavelength, µa.lx/ is the absorption coefficient of thefluorophore at the excitation wavelength and f.lm/ is itsfluorescence quantum yield at the emission wavelength.

Using the analytical method developed by Gardneret al., one can recover the intrinsic fluorescence coefficientfrom the measured fluorescence using Equation (11):

b.lx, lm/ D F.lx, lm/

P0.lx/[.�/p/ cos q]D.lm/X1D.lx, lm/.11/

where F.lx, lm/ is the measured fluorescence inten-sity, P0.lx/ is the power of the incident light at theexcitation wavelength, D is the detector’s wavelength-dependent response function, [.�/p/ cos q] is the detec-tor collection efficiency for a distant detector geometryand a tissue surface with Lambertian intensity distri-bution, and X1D is the one-dimensional path lengthfactor. The term P0.lx/[.�/p/ cos q]D.lm/ can be quan-tified by calibration of the fluorescence measurementfor source and detector constants. Specifically, the fluo-rescence calibration can be performed with an opticallydilute fluorophore solution with a predetermined intrin-sic fluorescence coefficient. The path length factor X1D

requires knowledge of the tissue optical properties atthe excitation and emission wavelengths. This functionis based upon the exponential attenuation of light awayfrom the source, as are the Beer–Lambert and diffusiontheories, with coefficients that are empirically developedfrom extensive Monte-Carlo simulations. Factor X1D canbe calculated from noninvasive measurements of the dif-fuse reflectance (Rd) and effective penetration depth (d).Because two optical properties are needed, it is possibleto make two, unique, diffuse reflectance measurementswith two illumination–collection fiber distances in orderto specify Rd and d.

The recovery method presented successfully identi-fies the intrinsic fluorescence coefficient (both spectralline shape and intensity) of turbid tissue phantoms thatincorporate various concentrations of fluorophore (rho-damine 6G), absorber (adult hemoglobin) and scatterer(polystyrene spheres) with absorption and scattering coef-ficients that coincide with the range found in soft tissues.The results are shown in Figure 19. Furthermore, fluo-rophore concentrations of the turbid tissue phantoms arepredicted to within 15% of the true concentration (notshown).

Although this analytical model is simple and canrecover the intrinsic fluorescence coefficient from turbidtissue phantom fluorescence emission spectra, there areseveral limitations associated with it. The optical prop-erties of the medium need to be uniform. Furthermore,the size of the medium should be large enough such thatonly the surface boundary affects the distribution of light.Finally, the one-dimensional fluence rate expression isvalid for collimated incident light with a beam diameterthat is at least a factor of four larger than the penetrationdepth of the light.


102

103

104

105

106 Phantom 1Phantom 2Phantom 3Phantom 4Phantom 5Phantom 6

(a)

Mea

sure

d flu

ores

cenc

e,F

(de

tect

or u

nits

) R6G/PBS

R6G/ALB/PBS

50010−5

10−4

10−3

10−2

550 600 650 700

Wavelength (nm)(b)

Intr

insi

c flu

ores

cenc

eco

effic

ient

, β (

cm−1

) R6G/PBS

R6G/ALB/PBS

Phantom 1Phantom 2Phantom 3Phantom 4Phantom 5Phantom 6

Figure 19 Summary of intrinsic fluorescence recovery fromsix turbid tissue phantoms: (a) measured uncorrected fluores-cence spectra; (b) corrected fluorescence spectra, displayed asthe intrinsic fluorescence coefficient b. Note that R6G/PBScorresponds to the fluorescence of rhodamine 6G in phos-phate-buffered saline and that R6G/ALB/PBS corresponds tothe fluorescence of rhodamine 6G in 2.5% bovine serum albu-min–phosphate-buffered saline. (Reproduced by permission ofthe Optical Society of America, from Gardner et al..127/)

Wu et al..126/ have achieved a similar goal as Gardneret al..127/ using a photon migration approach to modelfluorescence from a homogeneous turbid medium suchas tissue. The model provides an analytical relation-ship between the bulk fluorescence emission spectrumF and the diffuse reflectance spectrum R for arbitrarygeometries and boundary conditions, which representsan advantage over the previous analytical model..126/ Wuet al. demonstrate that the distortion can be removedby measuring R from a turbid medium over the samewavelength range and with the same geometry as isused for measuring F. The validity of this approachhas been demonstrated from tissue experiments usinghuman aortic media and with Monte-Carlo simulations.This analytical model does provide accurate fluorescenceline shape information, but unlike the previous analyt-ical model.126/ the absolute intensity is still coupled tothe absorption coefficient. This implies that the fluo-rescence measurements from two tissues that have the

same amount of fluorophore but different amounts ofabsorption will have the same line shape but differentfluorescence intensities.

Unlike Gardner et al..127/ and Wu et al.,.126/ Panou-Diamandi et al..145/ modeled tissue fluorescence usingelectromagnetic theory. In this mathematical approach,the inelastic property of fluorescence was expressedvia the polarization vector of the medium. Thefluorescence scattering spectral function was indepen-dent of the excitation and emission geometry, express-ing the energy transfer from the excitation frequencyto all the emission frequencies. The model assumedthat the tissue is a single, homogeneous, infinitelythick dielectric layer under plane wave illumination.Experimental measurements were carried out on opti-cally turbid collagen gels that contained fluorescentdyes, in order to validate the mathematical model.Comparison between experimental and theoreticallyexpected fluorescence emission spectra gave satisfactoryresults.

Another technique that has been employed to decon-volve the absorption and scattering properties from thefluorescence emission spectra of turbid media is par-tial least squares (PLS) regression..146/ This method isnot based on analytical approximations of the trans-port equations.126,127/ or Maxwell’s equations,.145/ butrather describes the fluorescence emission spectra mea-sured from turbid medium as a linear combinationof basis vectors that are representative of those tobe predicted. The PLS method involves the regres-sion between a fluorescence spectral matrix X.nðm/and a concentration matrix Y.nð p/. PLS seeks acalibration matrix B.mð p/ such that Equation (12)holds:

Y D XB .12/

The PLS method was compared to models of lighttransport based on Beer’s law.147/ and Kubelka–Munktheory.56/ to determine which is most effective inextracting the concentration of fluorophores in a setof turbid tissue phantoms containing absorbers andscatterers. The model based on Beer’s law significantlyunderestimated the fluorophore concentrations, whereasthe model based on Kubelka–Munk theory significantlyoverestimated the concentrations. However, this methodoverestimated the concentrations by a constant scalingfactor that potentially can be corrected for. Finally, usingthe PLS method, which was trained on a set of turbid tissuephantom spectra that have optical properties similar tothose to be analyzed, fluorophore concentrations werepredicted to within 5% of the true concentration. Itshould be noted that the success of the PLS method relieson a training set that reflects the chemical and optical


complexity of the turbid media to be investigated. Severalimportant obstacles remain before PLS can be appliedto analyze fluorescence emission spectra from tissues.The main obstacle is the question of how to constructreliably a training set in which the concentrations offluorophores and absorbers of biological interest can bequantified.

7.2.2 Identity and Distribution of Fluorophores thatContribute to Tissue Fluorescence Emission Spectra

It was mentioned earlier that quantification of theconcentration of fluorophores within tissue in principleinvolves two steps. The first step requires deconvolutionof the absorption and scattering properties of the tissuefrom the bulk tissue fluorescence emission spectrum. Thesecond step requires knowledge of the identity and dis-tribution of the fluorophores within the tissue. Althoughthe previously described analytical models.126,127,145,146/

address the first step, they do not address the second.The following example addresses the second step specifi-cally through the measurement and quantification of thebiochemical and morphological basis of colonic tissuefluorescence emission spectra measured in vivo..52/

Zonios et al..52/ measured and quantified the con-tribution of the fluorescent microstructures in frozen,unstained tissue sections to the tissue fluorescence emis-sion spectra measured from colonic tissues in vivo. Inorder to achieve this, first fluorescence emission spec-tra were measured at 370 nm excitation from normalcolon tissues and colonic adenomas (precancer) in vivo.Deconvolution of absorption and scattering from flu-orescence emission spectra measured from the tissuesin vivo was achieved using optical properties obtainedfrom Kubelka–Munk.61/ analysis of diffuse transmit-tance and reflectance measurements of excised humancolon tissues in vitro. Fluorescence, microscopy andmicrospectrofluorimetry at 363 nm excitation of thin,frozen, unstained tissue sections were used to charac-terize the spectral line shapes and the distribution of thefluorescent microstructures that contribute to the mea-sured fluorescence emission spectra of colonic tissues invivo. A model based on Monte-Carlo simulations.65/ wasused to relate the spectral line shape and distributionof the fluorescent microstructures to the colonic tissuespectra measured in vivo. This is perhaps one of the firstcomprehensive attempts to measure and quantify the bio-chemical and morphological basis of tissue fluorescenceemission spectra measured in vivo.

The model developed by Zonios et al..52/ describes thetissue fluorescence emission spectrum F.lx, lm/, as shownby Equation (13):

F.lx, lm/ D k∑

i

∫fi.lm/Di.z/T.lx, lm, z/ dz .13/

where lx represents the excitation wavelength, lm

represents the emission wavelength, z represents thedepth within the tissue sample, fi represents the intrinsicfluorescence of each individual microstructure with indexi, Di.z/ represents the fluorescence intensity spatialdistribution of an individual microstructure with indexi, T.lx, lm, z/ is a transfer function that incorporatesthe tissue optical properties and is calculated usingMonte-Carlo simulations of light propagation in tissue,and k represents a scaling factor.

First, a database of fluorescence line shapes and distri-butions of fluorescent microstructures within colon tissuewere obtained. A microspectrofluorimeter incorporat-ing a fluorescence microscope, an argon laser source(363.8 nm), a spectrograph and an intensified PDA,coupled to an optical multichannel analyzer was used torecord the fluorescence emission spectra of microstruc-tures from unstained, frozen tissue sections cut fromtissue biopsies. Quantitative fluorescence imaging studiesto quantify the distribution of the fluorescent microstruc-tures within the tissue section were performed using a75-W short-arc filtered xenon lamp (380 nm) coupled to amicroscope and a thermoelectrically cooled CCD camera.

Finally, the transmission and reflection spectra of thicksamples of normal mucosa and adenomatous polypswere measured in the range 300–700 nm using a spec-trophotometer equipped with an integrating sphere. Thetissue absorption and scattering properties were deter-mined by calculating the Kubelka–Munk coefficients.61/

and transforming these into transport theory absorptionand reduced scattering coefficients..148/ The anisotropyparameter was obtained from the literature for tissueswith histology similar to the colon..53/ The transfer func-tion was determined using Monte-Carlo simulations..65/

The transfer function accounts for the scattering andabsorption of the excitation light and emitted light atwavelengths lx and lm, respectively. It also accounts forthe specific light illumination/collection geometry used tomake the fluorescence measurements from tissues in vivo.

Figure 20 shows the computed fluorescence emissionspectra of normal and adenomatous colon tissue versusthe corresponding average fluorescence emission spectrameasured from tissues in vivo. The peak intensity ofthe in vivo spectra has been scaled for comparison. Thecomputed spectra contain all the characteristic spectralfeatures observed in the in vivo spectra. The followinginterpretations may be made from the intrinsic fluorescentmicrostructures, the fluorescence density function andthe transfer function used to calculate the computedfluorescence emission spectra shown in Figure 20:

ž Although the intrinsic fluorescence of collagen (thedominant fluorophore) peaks at 420 nm, the observedpeak in both normal tissues and adenoma is at 460 nm


0400 500 600 700

300

200

100

Wavelength (nm)

Flu

ores

cenc

e in

tens

ity (

a.u.

)

AdenomaNormalClinical data

Figure 20 Computed fluorescence emission spectra from nor-mal colon tissues and adenomas versus the average fluorescencespectra measured in vivo. (Reproduced by permission of IEEE,Piscataway, NJ ( 1996 IEEE), from Zonios et al..82/)

due to the effect of large hemoglobin absorption at420 nm.

ž The fluorescence intensity of adenoma is smaller thanthat of normal colon. This is due to three factors: themucosal collagen fluorescence is decreased in ade-noma due to the enlargement of crypts, which displacethe lamina propria; the submucosa contributes tothe fluorescence in normal tissue but not in ade-noma, because of the increased thickness of a polyp;and adenoma exhibits increased absorption due tohemoglobin content.

ž Red fluorescence is increased in adenoma. Thisadditional red fluorescence is primarily associatedwith the intrinsic fluorescence of the dysplastic cryptcells.

In conclusion, the model developed by Zonios et al..52/

quantifies the contribution of the fluorescent microstruc-tures that are responsible for differences observed inthe fluorescence emission spectra of normal and adeno-matous colon tissues. Their findings indicate that boththe biochemical and morphological features of the tis-sue contribute to the measured fluorescence emissionspectrum.

Zeng et al..149/ have used a similar approach toreconstruct in vivo skin fluorescence emission spectra.Specifically, they measured the fluorescence emissionand diffuse reflectance spectra of skin in vivo and thefluorescence of the microstructures in the skin in vitro.They reconstructed the spectra measured from skin invivo using Monte-Carlo simulations in which they usedthe microscopic fluorophore distribution and spectrameasured from excised skin tissue sections in vitro.The optical properties of skin, however, were obtainedfrom published values in the literature. They did not

report similar data for any skin lesions such as naevi ormelanoma.

7.3 Effect of Excitation and Emission Geometry onFluorescence Emission Spectra of Turbid Media

Several groups have evaluated the effect of the excitationand emission geometry on the measurement of tissuefluorescence emission spectra from turbid media andtissue..58,150 – 153/ Keijzer et al..58/ used Monte-Carlo simu-lations and experimental measurements to show that theillumination/collection geometry affects both the inten-sity and line shape of fluorescence emission measuredfrom tissue. Specifically, they demonstrated that as theseparation between the illumination and collection fiber isincreased, the measured fluorescence intensity decreasesas a result of hemoglobin absorption. Furthermore, theattenuation due to hemoglobin absorption is wavelengthdependent, thus affecting the fluorescence line shape aswell. Richards-Kortum et al..57/ suggested using a fiber-optic probe, which collects the emitted light from an area,that is directly illuminated by the excitation light in orderto minimize the effect of hemoglobin absorption on thefluorescence measurement. Qu et al..150/ also used Monte-Carlo simulations and experimental measurements todemonstrate that the distortion to the fluorescence emis-sion spectrum is a function of the diameter of the areaviewed by the illumination and collection fiber as well.Using bronchial tissue optical properties (for the excita-tion and emitted light) in their Monte-Carlo simulations,they found that the distortions to the fluorescence emis-sion spectrum were minimized when the diameter of thetissue area viewed by the excitation and emission fiberwas less than 1 mm. This was verified experimentally.

Pogue et al..151/ went a step further and demonstratedthat if fluorescence is measured from a sample volumethat is smaller than the average mean free scattering path(inverse of the reduced scattering coefficient) within thatturbid medium, then the effects of absorption will bediminished. To achieve this, they used a confocal detec-tion scheme to measure fluorescence as a function offluorophore concentration from a turbid medium. Theyused a pinhole detector with a diameter (10 µm) that wasan order of magnitude smaller than the mean free scatter-ing path in the medium. Using this method, they were ableto obtain a linear relationship between the fluorescenceintensity and concentration of the fluorophore in theturbid medium for a wide range of fluorophore concentra-tions. Although this relationship was independent of theabsorption coefficient, the fluorescence intensity variedlinearly with the scattering coefficient for a fixed fluo-rophore concentration. Fortuitously, this is not a seriouslimitation because most tissues do not have a large vari-ation in their scattering coefficient. It was also observed


for a fixed detector size that, as the mean free scatteringpath decreases, the number of nonscattered fluorescentphotons relative to the number of scattered fluorescentphotons that are detected also decreases. This suggeststhat the size of the detector has to be reduced as themean free scattering path length of the turbid medium isreduced. This has direct implications on the strength ofthe fluorescence intensities detected and the size of thesampled volume. In a subsequent report, Pogue et al..152/

showed that they could overcome the problem of poorsignal-to-noise ratio by developing and employing a fiber-optic probe that has multiple fibers (100 µm in diameter),each of which serves as an independent source and detec-tor. Using this probe, the fluorescence intensity measuredby each fiber was obtained from a sample volume smallerthan the mean free scattering path for that medium,thus minimizing the effect of absorption. By integratingthe fluorescence intensity from a total of 30 fibers, thesignal-to-noise ratio of the fluorescence measurementswas enhanced significantly. The main limitation of thistechnique is the very small sampled volume.

Avrillier et al..153/ used an alternative approach toresolve the effect of the illumination/collection geometryon the fluorescence emission spectra of tissue. Theyused Monte-Carlo simulations and optical propertiesof tissues measured in vitro to predict the effect ofabsorption and scattering on the tissue fluorescenceemission spectra measured in vivo. Specifically, theypredicted the correction factors as a function of increasedseparation between the excitation and emission fibers.These predictions were used to correct the distortionin the fluorescence emission spectra measured from thetissues. However, because the optical properties of tissuesmeasured in vitro are fraught with error associated withextraction, this approach is considered to be, at best,approximate unless the tissue optical properties used inthe Monte-Carlo simulations are measured in vivo.

8 FUTURE PERSPECTIVES

Diagnostic tools based on optical spectroscopy in theUV/VIS spectral regions have the potential to link thebiochemical and morphological properties of tissues toindividual patient care. In particular, these techniquesare fast, noninvasive and quantitative. Furthermore,the accuracy and efficacy of the technology have beendemonstrated clinically to be comparable or superior tocurrent clinical modalities.

However, the biochemical and morphological basis forthe diagnostic capability of fluorescence spectroscopy isnot completely understood. This is limited by the lack of:sophisticated mathematical models that can be used toquantify tissue optical properties in order to account for

the effects of absorption and scattering in the UV/VISspectral region; and a comprehensive understandingof the biochemical constituents that contribute to themeasured fluorescence. Hence, today’s challenge forbiomedical optics is to elucidate tissue biochemistry andmorphology in greater detail for the specific diseaseprocess so that appropriate therapeutic interventionsmay have the greatest impact. In order to achieve this,both experimental and mathematical techniques need todeveloped that can enable a greater understanding of thechromophores and fluorophores that are modulated bythe disease process and their contribution to the measuredspectrum.

Animal models are useful with regard to developingmethodologies that can be used to elucidate the biochem-ical and morphological properties of tissue and optimizingthe instruments to exploit these features. The advantagesthat they offer are: they are well characterized; spectralmeasurements can be made per tissue site for a varietyof experimental and biological conditions; and data canbe obtained from a statistically significant number of ani-mals for the purpose of validation without the need forexpensive clinical trials. Although animal model studieshave been performed extensively in relation to the use ofphotosensitizing agents for PDT, only a handful of stud-ies on measuring and quantifying tissue autofluorescencehave been performed..154 – 159/

With respect to clinical applications, the diagnosticpotential of fluorescence spectroscopy is defined by his-tology. In order to evaluate if fluorescence spectroscopyis sensitive to changes that precede the morphologicalmanifestations seen by histological evaluation, it is valu-able to correlate the measured spectral information toother biochemical and/or genetic markers of increasedcancer risk. Furthermore, although fluorescence spec-troscopy has been most widely explored for the detectionof precancer or cancer in a screening and diagnostic set-ting, its diagnostic potential for guiding surgery has beenevaluated only in a limited number of clinical studies..101/

Because this technique can discriminate between dis-eased and nondiseased tissues in a fast and noninvasivemanner, it could be used to define margins of resectionfor various surgical procedures, such as that of the breastand brain. Additionally, this technique also could be usedto monitor the response of tissues to various therapeuticinterventions.

ACKNOWLEDGMENTS

The author gratefully acknowledges Ms Mary Leonard forher extensive help with the preparation of this manuscript.


ABBREVIATIONS AND ACRONYMS

APD Avalanche PhotodiodeATP Adenosine TriphosphateBPD Benzoporphyrin DerivativeCCD Charge-coupled DeviceEEM Excitation–Emission MatrixFAD Flavin Adenine DinucleotideFADH2 Reduced Flavin Adenine DinucleotideFp Flavoproteinfwhm Full Width at Half-maximumHpD Hematoporphyrin DerivativeIR InfraredLIFE Light-induced Fluorescence EndoscopyMTHPC Meso-tetra-(hydroxyphenyl)-chlorinNADC Nicotinamide Adenine DinucleotideNADH Reduced Nicotinamide Adenine

DinucleotideNIR Near-infraredNIST National Institute of Standards

and TechnologyPCA Principal Component AnalysisPDA Photodiode ArrayPDT Photodynamic TherapyPLS Partial Least SquaresPMT Photomultiplier TubePN Pyridine NucleotidePpIX Protoporphyrin IXSIL Squamous Intra-epithelial LesionsSnET2 Tin EtiopurpurinUV UltravioletVIS Visible5-ALA d-Aminolevulinic Acid

RELATED ARTICLES

Biomedical Spectroscopy (Volume 1)Biomedical Spectroscopy: Introduction ž Infrared Spec-troscopy in Clinical and Diagnostic Analysis ž InfraredSpectroscopy, Ex Vivo Tissue Analysis by žNear-infraredSpectroscopy, In Vivo Tissue Analysis by žPhotodynamicTherapy

REFERENCES

1. R.S. Cotran, V. Kumar, S.L. Robbins, Robbins Patho-logic Basis of Disease, WB Saunders Co., Philadelphia,PA, 1989.

2. I.D. Campbell, R.A. Dwek, Biological Spectroscopy,Benjamin/Cummings Publishing Co., Menlo Park, CA,1984.

3. F. Koenig, F.J. McGovern, H. Enquist, R. Larne, T.F.Deutsch, K.T. Schomacker, ‘Autofluorescence GuidedBiopsy for the Early Diagnosis of Bladder Carcinoma’,J. Urol., 159(6), 1871–1875 (1998).

4. B. Chance, M. Cope, B. Dugan, N. Ramanujam,B. Tromberg, ‘Phase Measurements of Absorbers andScatterers in Tissue: Review’, Rev. Sci. Instrum., 69(10),3457–3481 (1998).

5. S. Andersson-Engels, J. Johansson, U. Stenram, K. Svan-berg, S. Svanberg, ‘Malignant Tumor and Atheroscle-rotic Plaque Diagnosis Using Laser Induced Fluores-cence’, IEEE J. Quant. Electron., 26(12), 2207–2217(1990).

6. S. Andersson-Engels, J. Johansson, K. Svanberg, S.Svanberg, ‘Fluorescence Imaging and Point Measure-ments of Tissue: Applications to the Demarcation ofMalignant Tumors and Atherosclerotic Lesions fromNormal Tissue’, Photochem. Photobiol., 53(6), 807–814(1991).

7. S. Andersson-Engels, B.C. Wilson, ‘In Vivo Fluores-cence in Clinical Oncology: Fundamentals and PracticalIssues’, J. Cell. Pharmacol., 3, 66–79 (1992).

8. T.G. Papazoglou, ‘Malignancies and AtheroscleroticPlaque Diagnosis – Is Laser Induced Fluorescence Spec-troscopy the Ultimate Solution?’, J. Photochem. Photo-biol. B: Biol., 28, 3–11 (1995).

9. I.J. Bigio, J.R. Mourant, ‘Ultraviolet and Visible Spec-troscopies for Tissue Diagnostics: Fluorescence Spec-troscopy and Elastic-scattering Spectroscopy’, Phys.Med. Biol., 42, 803–814 (1996).

10. S. Andersson-Engels, C.A. Klinteberg, K. Svanberg,S. Svanberg, ‘In Vivo Fluorescence Imaging forTissue Diagnostics’, Phys. Med. Biol., 42, 815–824(1997).

11. R. Richards-Kortum, E. Sevick-Muraca, ‘QuantitativeOptical Spectroscopy for Tissue Diagnosis’, Annu. Rev.Phys. Chem., 47, 555–606 (1996).

12. G.A. Wagnieres, W.M. Star, B.C. Wilson, ‘In Vivo Flu-orescence Spectroscopy and Imaging for OncologicalApplications’, Photochem. Photobiol., 68(5), 603–632(1998).

13. E.B. Hanlon, I. Itzkan, R.R. Dasari, M.S. Feld, R.J. Fer-rante, A.C. McKee, D. Lathi, N.W. Kowall, ‘Near-infrared Fluorescence Spectroscopy Detects Alzheimer’sDisease In Vitro’, Photochem. Photobiol., 70(2), 236–242(1999).

14. L.T. Perelman, V. Backman, M. Wallace, G. Zonios,R. Manoharan, A. Nusrat, S. Shields, M. Seiler, C. Lima,T. Hamano, I. Itzkan, J. Van Dam, J.M. Crawford,M.S. Feld, ‘Observation of Periodic Fine Structure inReflectance from Biological Tissue: a New Techniquefor Measuring Nuclear Size Distribution’, Phys. Rev.Lett., 80(3), 627–630 (1998).

15. B.B. Das, F. Liu, R.R. Alfano, ‘Time-resolved Fluores-cence and Photon Migration Studies in Biomedical


and Random Media’, Rep. Prog. Phys., 60(2), 227–292(1997).

16. S.P. Gopinath, C.S. Robertson, R.G. Grossman, B. Cha-nce, ‘Near-infrared Spectroscopic Localization of Intra-cranial Hematomas’, J. Neurosurg., 79, 43–47 (1993).

17. W.J. Levy, S. Levin, B. Chance, ‘Near-infrared Mea-surement of Cerebral Oxygenation: Correlation withElectroencephalographic Ischemia During VentricularFibrillation’, Anesthesiology, 83(4), 738–746 (1995).

18. S. Nioka, Q. Luo, B. Chance, ‘Human Brain FunctionalImaging with Reflectance CWS’, Adv. Exp. Med. Biol.,428, 237–242 (1997).

19. A. Villringer, B. Chance, ‘Non-invasive Optical Spec-troscopy and Imaging of Human Brain Function’, TrendsNeurosci., 20(10), 435–442 (1997).

20. W. Bank, B. Chance, ‘Diagnosis of Defects in OxidativeMuscle Metabolism by Non-invasive Tissue Oximetry’,Mol. Cell. Biochem., 174(1/2), 7–10 (1997).

21. S. Nioka, M. Miwa, S. Orel, M. Shnall, M. Haida,S. Zhao, B. Chance, ‘Optical Imaging of Human BreastCancer’, Adv. Exp. Med. Biol., 361, 171–179 (1994).

22. D.A. Boas, A.G. Yodh, ‘Spatially Varying DynamicalProperties of Turbid Media Probed with DiffusingTemporal Light Correlation’, J. Opt. Soc. Am. A, 14,192–215 (1997).

23. A. Mahadevan-Jansen, R. Richards-Kortum, ‘RamanSpectroscopy of Pre-cancers and Cancers’, J. Biomed.Opt., 1(1), 31–70 (1996).

24. J.F. Brennan III, M.E. Beattie, Y. Wang, M.J. Cantella,B.Y. Tsaur, R.R. Dasari, M.S. Feld, ‘PdSi Focal-planeArray Detectors for Short-wave Infrared Raman Spec-troscopy of Biological Tissue: a Feasibility Study’, Appl.Opt., 35(28), 5736–5739 (1996).

25. A.J. Berger, T.W. Koo, I. Itzkan, G. Horowitz, M.S.Feld, ‘Multicomponent Blood Analysis by Near-infraredRaman Spectroscopy’, Appl. Opt., 38(13), 2916–2926(1999).

26. J.R. Lakowicz, Principles of Fluorescence Spectroscopy,Plenum Press, New York, 1983.

27. M.B. Silberberg, H.E. Savage, G.C. Tang, P.G. Sacks,R.R. Alfano, S.P. Schantz, ‘Detection of Retinoic AcidInduced Biochemical Alterations in Squamous CellCarcinoma Using Intrinsic Fluorescence Spectroscopy’,Laryngoscopy, 104(3), 278–282 (1994).

28. A. Pradhan, P. Pal, G. Durocher, L. Villeneuve, A. Ba-lassy, F. Babai, L. Gaboury, L. Blanchard, ‘SteadyState and Time Resolved Fluorescence Properties ofMetastatic and Non-metastatic Malignant Cells fromDifferent Species’, J. Photochem. Photobiol. B: Biol.,3(3), 101–112 (1995).

29. S. Ganesan, P.G. Sacks, Y. Yang, A. Katz, M. Al-Rawi,H.E. Savage, S.P. Schantz, R.R. Alfano, ‘Native Fluores-cence Spectroscopy of Normal and Malignant EpithelialCells’, Cancer Biochem. Biophys., 16, 365–373 (1998).

30. T.J. Romer, M. Fitzmaurice, R.M. Cothren, R. Richards-Kortum, M.V. Sivak, Jr, J.R. Kramer, Jr, ‘Laser-induced

Fluorescence Microscopy of Normal Colon and Dys-plasia in Colonic Adenomas: Implications for Spectro-scopic Diagnosis’, Am. J. Gastroenterol., 90(1), 81–87(1995).

31. H. Wang, J. Willis, M.J.F. Canto, M.V. Sivak, Jr, J.A.Izatt, ‘Quantitative Laser Scanning Confocal Autoflu-orescence Microscopy of Normal, Premalignant, andMalignant Colonic Tissues’, IEEE Trans. Biomed. Eng.,46(10), 1246–1252 (1999).

32. G.S. Fairman, M.H. Nathanson, A.B. West, L.I. Deckel-baum, L. Kelly, C.R. Kapadia, ‘Differences in Laser-induced Autofluorescence Between Adenomatous andHyperplastic Polyps and Normal Colonic Mucosaby Confocal Microscopy’, Digest. Dis. Sci., 40(6),1261–1268 (1995).

33. K. Izuishi, H. Tajiri, T. Fujii, N. Boku, A. Ohtsu,T. Ohnishi, M. Ryi, T. Kinoshita, S. Yoshida, ‘TheHistological Basis of Detection of Adenoma and Cancerin the Colon by Autofluorescence Endoscopic Imaging’,Endoscopy, 31(7), 511–516 (1999).

34. G. Bottiroli, A.C. Croce, D. Locatelli, R. Marchesini,E. Pignoli, S. Tomatis, C. Cuzzoni, S. Di Palma, M. Dal-fante, P. Spinellu, ‘Natural Fluorescence of Normal andNeoplastic Human Colon: a Comprehensive ‘‘ex vivo’’Study’, Lasers Surg. Med., 16(1), 48–60 (1995).

35. G.I. Zonios, R.M. Cothren, J.T. Arendt, J. Wu, J. VanDam, J.M. Crawford, R. Manoharan, M.S. Feld, ‘Mor-phological Model of Human Colon Tissue Fluores-cence’, IEEE Trans. Biomed. Eng., 43(2), 113–122(1996).

36. R.S. DaCosta, L. Ligle, J. Kost et al., ‘Confocal Micros-copy/Macroscopy and Microspectrofluorimetry Analysisof Human Colorectal Tissue’, J. Anal. Morphol., 4,192–194 (1997).

37. W.S. Glassman, M. Steinberg, R.R. Alfano, ‘Time Re-solved and Steady State Fluorescence Spectroscopy fromNormal and Malignant Cultured Human Breast CellLines’, Lasers Life Sci., 6(2), 91–98 (1994).

38. M. Anidjar, O. Cussenot, J. Blais, O. Bourdon, S. Avril-lier, D. Ettori, J.M. Villter, J. Fiet, P. Teillac, A. Le Duc,‘Argon Laser Induced Autofluorescence may Distin-guish Between Normal and Tumor Human UrothelialCells: a Microspectrofluorimetric Study’, J. Urol., 155(5),1771–1774 (1996).

39. F.N. Ghadially, W.J.P. Neish, H.C. Dawkins, ‘Mecha-nisms Involved in the Production of Red Fluorescence ofHuman and Experimental Tumors’, J. Pathol. Bacteriol.,85, 77–92 (1963).

40. L. Stryer, Biochemistry, WH Freeman and Company,New York, 1988.

41. D. Fujimoto, ‘The Structure of Pyridinoline, a CollagenCrosslink’, Biochem. Biophys. Res. Commun., 76(4),1124–1129 (1977).

42. Z. Deyl, K. Macek, M. Adam, O. VanCikova, ‘Studieson the Chemical Nature of Elastin Fluorescence’,Biochim. Biophys. Acta, 625, 248–254 (1980).


43. D.R. Eyre, M.A. Paz, ‘Cross-linking in Collagen andElastin’, Annu. Rev. Biochem., 53, 717–748 (1984).

44. D.P. Thornhill, ‘Separation of a Series of Chromophoresand Fluorophores Present in Elastin’, Biochem. J., 147,215–219 (1975).

45. B. Chance, B. Schoener, R. Oshino, F. Itshak, Y. Nakase,‘Oxidation–Reduction Ratio Studies of Mitochondriain Freeze-trapped Samples’, J. Biol. Chem., 254(11),4764–4771 (1979).

46. A.M. Kluftinger, N.L. Davis, N.F. Quenville, S. Lam,J. Hung, B. Palcic, ‘Detection of Squamous Cell Cancerand Pre-cancerous Lesions by Imaging Tissue Autoflu-orescence in the Hamster Cheek Pouch Model’, Surg.Oncol., 1, 183–188 (1992).

47. A. Policard, ‘Etudes sur les Aspects Offerts par desTumeurs Experimentales Examinees a la Lumiere deWoods’, CR Soc. Biol., 91, 1423–1425 (1924).

48. F. Ronchese, ‘The Fluorescence of Cancer Under theWood’s Light’, Oral Surg. Oral Med. Pathol., 7, 967–971(1954).

49. F.N. Ghadially, ‘Red Fluorescence of ExperimentallyInduced and Human Tumors’, J. Pathol. Bacteriol., 80,345–361 (1960).

50. F.N. Ghadially, W.J.P. Neish, ‘Porphyrin Fluorescenceof Experimentally Induced Squamous Cell Carcinoma’,Nature, 188, 1124 (1960).

51. F.N. Ghadially, W.J.P. Neish, H.C. Dawkins, ‘Mecha-nisms Involved in the Production of Red Fluorescence ofHuman and Experimental Tumors’, J. Pathol. Bacteriol.,85, 77–92 (1963).

52. J.C. Kennedy, R.H. Pottier, D.C. Pross, ‘PhotodynamicTherapy with Endogenous Protoporphyrin IX: BasicPrinciples and Present Clinical Experience’, J. Pho-tochem. Photobiol. B, 6, 143–148 (1990).

53. W.F. Cheong, A.J. Welch, ‘A Review of the OpticalProperties of Tissues’, IEEE J. Quantum Electron., 26,2166–2185 (1990).

54. R. Richards-Kortum, ‘Fluorescence Spectroscopy ofTurbid Media’, in Optical–Thermal Response of LaserIrradiated Tissue, eds. A.J. Welch, M.J.C. van Gemert,Plenum Press, New York, 667–706, 1995.

55. S.B. Brown, An Introduction to Spectroscopy for Bio-chemists, Academic Press, London, 1980.

56. A.J. Durkin, S. Jaikumar, N. Ramanujam, R. Richards-Kortum, ‘Relation Between Fluorescence Spectra ofDilute and Turbid Samples’, Appl. Opt., 33(3), 414(1994).

57. R. Richards-Kortum, A. Mehta, G. Hayes, R. Cothren,T. Kolubayev, C. Kitrell, N.B. Ratliff, J.R. Kramer, M.S.Feld, ‘Spectral Diagnosis of Atherosclerosis Using anOptical Fiber Laser Catheter’, Am. Heart J., 118(2), 381(1989).

58. M. Keijzer, R. Richards-Kortum, S.L. Jacques, M.S.Feld, ‘Fluorescence Spectroscopy of Turbid Media:Autofluorescence of the Human Aorta’, Appl. Opt., 28,4286–4292 (1989).

59. S. Chandrasekhar, Radiative Transfer, Oxford Univer-sity Press, Oxford, 1950.

60. A. Ishimaru, Wave Propagation and Scattering in Ran-dom Media, Academic Press, New York, 1987.

61. P. Kubelka, ‘New Contributions to the Optics ofIntensely Light Scattering Materials. Part II: Non-homogeneous Layers’, J. Opt. Soc. Am., 344, 330–335(1954).

62. S. Prahl, M.J.C. van Gemert, A.J. Welch, ‘Determiningthe Optical Properties of Turbid Media by Using theAdding–Doubling Method’, Appl. Opt., 32, 559–568(1993).

63. M.S. Patterson, B. Chance, B.C. Wilson, ‘Time ResolvedReflectance and Transmittance for the Non-invasiveMeasurement of Tissue Optical Properties’, Appl. Opt.,28, 2331–2336 (1989).

64. J.J. Duderstadt, L.J. Hamilton, Nuclear Reactor Analy-sis, Wiley, New York, 103–144, 1976.

65. S.L. Jacques, ‘Monte-Carlo Methods’, in Optical–Ther-mal Response of Laser-irradiated Tissue, eds. A.J. Welch,M.J.C. van Gemert, Plenum Press, New York, 73–99,1995.

66. A.J. Durkin, S. Jaikumar, R. Richards-Kortum, ‘Opti-cally Dilute, Absorbing and Turbid Phantoms forFluorescence Spectroscopy of Homogeneous and Inho-mogeneous Samples’, Appl. Spectrosc., 47, 2114–2121(1993).

67. G. Wagnieres, S. Cheng, M. Zellweger, N. Utke,D. Braichotte, J.-P. Ballini, H. van den Bergh, ‘AnOptical Phantom with Tissue-like Properties in theVisible for Use in PDT and Fluorescence Spectroscopy’,Phys. Med. Biol., 42(7), 1415–1426 (1997).

68. E. Kohen, R. Santus, J.G. Hirschberg, Photobiology,Academic Press, San Diego, CA, 1995.

69. R.A. Zangaro, L. Silveira, R. Manoharan, G. Zonios,I. Itzkan, R.R. Dasari, J. Van Dam, M.S. Feld, ‘RapidMulti-excitation Fluorescence Spectroscopy System forIn Vivo Tissue Diagnosis’, Appl. Opt., 35(25), 5211–5219(1996).

70. R.L. McCreery, ‘CCD Array Detectors for Multi-channel Raman Spectroscopy’, in Charged TransferDevices in Spectroscopy, eds. J.V. Sweedler, K.L. Ratz-laff, M.B. Denton, VCH Publishers, New York, 227–279,1994.

71. E.V. Trujillo, D.R. Sandison, U. Utzinger, N. Ramanu-jam, M. Follen Mitchell, R. Richards-Kortum, ‘Methodto Determine Tissue Fluorescence Efficiency In Vivoand Predict Signal-to-noise Ratio for Spectrometers’,Appl. Spectrosc., 52(7), 943–951 (1998).

72. D.G. Taylor, J.N. Demas, Anal. Chem., 51, 712 (1979).73. G.G. Stokes, ‘Uber Die Anderung der Brechbarkeit des

Lichtes’, Philos. Trans. R. Soc., 107, 11 (1852).74. H. Stubel, ‘Die Fluoreszenz Tierischer Gewebe in

Ultravioletten Licht’, Pflugers Arch. Physiol., 142, 1(1911).

75. R.M. Lycette, R.B. Leslie, Lancet, 2, 436 (1965).


76. A.E. Profio Profio, D.R. Doiron, E.G. King, ‘Laser Flu-orescence Bronchoscope for Localization of Occult LungTumors’, Med. Phys., 6(6), 523–525 (1979).

77. A.E. Profio, D.R. Doiron, ‘A Feasibility Study of theUse of Fluorescence Bronchoscopy for the Localizationof Small Lung Tumors’, Phys. Med. Biol., 22, 949–957(1977).

78. R.R. Alfano, D.B. Tata, J. Cordero, P. Tomashefsky,F.W. Longo, M.A. Alfano, ‘Laser Induced Fluores-cence Spectroscopy from Native Cancerous and Nor-mal Tissue’, IEEE J. Quant. Electron., 20, 1507–1511(1984).

79. W. Lohmann, ‘In Situ Detection of Melanomas byFluorescence Measurements’, Maturwissenschaften, 75,201–202 (1988).

80. Y. Yang, Y. Ye, F. Li, Y. Li, P. Ma, ‘CharacteristicAutofluorescence for Cancer Diagnosis and its Origin’,Lasers Surg. Med., 7, 528–532 (1987).

81. A. Mahadevan, M.F. Mitchell, E. Silva, S. Thomsen,R. Richards-Kortum, ‘A Study of the Fluorescence Prop-erties of Normal and Neoplastic Human Cervical Tissue’,Lasers Surg. Med., 13, 647–655 (1994).

82. G.I. Zonios, R.M. Cothren, J.T. Arendt, J. Wu, J. VanDam, J.M. Crawford, R. Manoharan, M.S. Feld, ‘Mor-phological Model of Human Colon Tissue Fluo-rescence’, IEEE Trans. Biomed. Eng., 43, 113–122(1996).

83. K.T. Schomacker, J.K. Frisoli, C.C. Compton, T.J. Flotte,J.M. Richter, N.S. Nishioka, T.F. Deutsch, ‘UltravioletLaser-induced Fluorescence of Colonic Tissue: BasicBiology and Diagnostic Potential’, Lasers Surg. Med.,12(1), 63–78 (1992).

84. C. Eker, S. Montan, E. Jaramillo, K. Koizumi, C. Rubio,S. Andersson-Engels, K. Svanberg, S. Svanberg, P. Sle-zak, ‘Clinical Spectral Characterization of ColonicMucosal Lesions Using Autofluorescence and DeltaAmniolevulinic Acid Sensitization’, Gut, 44(4), 511–518(1999).

85. R.M. Cothren, M.V. Sivak, J. Van Dam, R.E. Petras,M. Fitzmaurice, J.M. Crawford, J. Wu, J.F. Brennan,R.P. Rava, R. Manoharan, M.S. Feld, ‘Detection of Dys-plasia at Colonoscopy Using Laser Induced Fluores-cence: a Blinded Study’, Gastrointest. Endosc., 44(2),168–176 (1996).

86. T.D. Wang, J.M. Crawford, M.S. Feld, Y. Wang, I. Itz-kan, J. Van Dam, ‘In Vivo Identification of ColonicDysplasia Using Fluorescence Endoscopic Imaging’,Gastrointest. Endosc., 49(4), 447–455 (1999).

87. M.A. Mycek, K.T. Schomacker, N.S. Nishioka, ‘ColonicPolyp Differentiation Using Time-resolved Autofluo-rescence Spectroscopy’, Gastrointest. Endosc., 48(4),390–394 (1998).

88. N. Ramanujam, M.F. Mitchell, A. Mahadevan-Jansen,S. Thomsen, G. Staerkel, A. Malpica, T. Wright,A. Atkinson, R. Richards-Kortum, ‘Cervical Pre-cancerDetection Using a Multivariate Statistical Algorithm

Based on Laser Induced Fluorescence Spectra at Multi-ple Excitation Wavelengths’, Photochem. Photobiol., 64,720–735 (1996).

89. K. Tumer, N. Ramanujam, J. Ghosh, R. Richards-Kortum, ‘Ensembles of Radial Basis Function Networksfor Spectroscopic Detection of Cervical Precancer’,IEEE Trans. Biomed. Eng., 45(8), 953–961 (1998).

90. S. Lam, T. Kennedy, M. Unger, Y.E. Miller, D. Gelmont,V. Rusch, B. Gipe, D. Howard, J. LeRiche, A. Coldman,A. Gazdar, ‘Localization of Bronchial IntraepithelialNeoplastic Lesions by Fluorescence Bronchoscopy’,Chest, 113(3), 696–702 (1998).

91. F. Koenig, F.J. McGovern, A.F. Althausen, T.F.Deutsch, K.T. Schomacker, ‘Laser Induced Autofluo-rescence Diagnosis of Bladder Cancer’, J. Urol., 156(5),1597–1601 (1996).

92. M. Anidjar, D. Ettori, O. Cussenot, P. Meria,F. Desgrandchamps, A. Cortesse, P. Teillac, A. Le Duc,S. Avrillier, ‘Laser Induced Autofluorescence of Blad-der Tumors: Dependence on Excitation Wavelength’, J.Urol., 156(5), 1590–1596 (1996).

93. M. Panjehpour, B.F. Overholt, T. Vo-Dinh, R.C. Haggit,D.H. Edwards, F.P. Buckely III, ‘Endoscopic Fluores-cence Detection of High Grade Dysplasia in Barrett’sEsophagus’, Gastroenterology, 111, 93–101 (1996).

94. J.K. Dhingra, D.F. Perrault, K. McMillan, E.E. Rebeiz,S. Kabani, R. Manoharan, I. Itzkan, M.S. Feld,S.M. Shapshay, ‘Early Diagnosis of Upper AerodigestiveTract Cancer by Autofluorescence’, Arch. Otolaryn-gol. – Head Neck Surg., 122(11), 1181–1186 (1996).

95. A. Gillenwater, R. Jacob, R. Ganeshappa, B. Kemp,A.K. El-Naggar, J.L. Palmer, G. Clayman, M.F. Mit-chell, R. Richards-Kortum, ‘Noninvasive Diagnosis ofOral Neoplasia Based on Fluorescence Spectroscopyand Native Tissue Autofluorescence’, Arch. Otolaryn-gol. – Head Neck Surg., 124(11), 1251–1258 (1998).

96. B. Kulapaditharom, V. Boonkitticharoan, ‘Laser Indu-ced Fluorescence Imaging in Localization of Head andNeck Cancers’, Ann. Otol. Rhinol. Laryngol., 107(3),241–246 (1998).

97. C.S. Betz, M. Mehlmann, K. Rick, H. Stepp, G. Grevers,R. Baumgartner, A. Leunig, ‘Autofluorescence Imagingand Spectroscopy of Normal and Malignant Mucosa inPatients with Head and Neck Cancer’, Lasers Surg. Med.,25, 323–334 (1999).

98. M.L. Harries, S. Lam, C. MacAulay, J.A. Qu, B. Palcic,‘Diagnostic Imaging of the Larynx: Autofluorescence ofLaryngeal Tumors Using the Helium Cadmium Laser’,J. Laryngol. Otol., 109, 108–110 (1995).

99. M. Zargi, L. Smid, I. Fajdiga, B. Bubnic, J. Lenarcic,P. Oblak, ‘Laser Induced Fluorescence in Diagnosticsof Laryngeal Cancer’, Acta Oto-Laryngol. Suppl., 527,125–127 (1997).

100. B.W. Chwirot, S. Chwirot, J. Redzinski, Z. Michinie-wicz, ‘Detection of Melanomas by Digital Imaging of


Spectrally Resolved Ultraviolet Light Induced Autoflu-orescence of Human’, Eur. J. Cancer, 34(11), 1730–1734(1998).

101. G. Bottiroli, A.C. Croce, D. Locatelli, R. Nano,E. Giombelli, A. Messina, E. Benericetti, ‘Brain TissueAutofluorescence: an Aid for Intra-operative Delin-eation of Tumor Resection Margins’, Cancer Detect.Prevent., 22(4), 330–339 (1998).

102. K. Izuishi, H. Tajiri, M. Ryu, J. Furuse, Y. Maru,K. Inoue, M. Konishi, T. Kinoshita, ‘Detection of BileDuct Cancer by Autofluorescence Cholangioscopy: aPilot Study’, Hepato-Gastroenterology, 46(26), 804–807(1999).

103. B.W. Chwirot, S. Chwirot, W. Jedrzejczyk, M. Jack-owski, A.M. Raczynska, J. Winczakiewicz, J. Dobber,‘Ultraviolet Laser Induced Fluorescence of HumanStomach Tissues: Detection of Cancer Tissues by Imag-ing Techniques’, Lasers Surg. Med., 21(2), 149–158(1997).

104. R.R. Alfano, G.C. Tang, A. Pradhan, W. Lam, D.S.J.Choy, E. Opher, ‘Fluorescence Spectra from Cancerousand Normal Breast and Lung Tissue’, IEEE J. Quant.Electron., 23, 1806–1811 (1987).

105. Y. Yang, A. Katz, E.J. Celmer, M. Zurawska-Szczepaniak, R.R. Alfano, ‘Optical Spectroscopy ofBenign and Malignant Tissues’, Lasers Life Sci., 7(2),115–127 (1996).

106. Y. Yang, A. Katz, E.J. Celmer, M. Zurawska-Szczepaniak, R.R. Alfano, ‘Excitation Spectrum ofMalignant and Benign Breast Tissues: a Potential Opti-cal Biopsy Approach’, Lasers Life Sci., 7(4), 249–265(1997).

107. P.K. Gupta, S.K. Majumder, A. Uppal, ‘Breast CancerDiagnosis Using Nitrogen Laser Excited Autofluores-cence Spectroscopy’, Lasers Surg. Med., 21(5), 417–422(1997).

108. H. Auler, G. Banzer, ‘Untersuchungen uber die Rolleder Porphine bei Geschwulstkranken Menschen undTieren’, Krebforschung, 53, 1423–1425 (1942).

109. R.H. Figge, ‘Near Ultraviolet Rays and FluorescencePhenomena as Aids to Discovery and Diagnosisin Medicine’, Univ. Md. Med. Bull., 26, 165–168(1942).

110. G.E. Moore, W.T. Peyton, L.A. French, W.W. Walker,‘The Clinical Use of Fluorescin in Neurosurgery’, J.Neurosurg., 5, 392–398 (1948).

111. G.C. Peck, H.P. Mack, H.F. Figge, ‘Cancer Detectionand Therapy III. Affinity of Lymphatic Tissues forHematoporphyrin’, Bull. School Med. Univ. Md., 38,124–127 (1953).

112. G.C. Peck, H.P. Mack, W.A. Holbrook, H.F. Figge, ‘Useof Hematoporphyrin Fluorescence in Biliary and CancerSurgery’, Bull. School Med. Univ. Md., 38, 124–127(1955).

113. D.S. Rassmussen-Taxdal, G.R. Ward, F.H. Figge, ‘Flu-orescence of Human Lymphatic and Cancer Tissue

Following High Doses of Intravenous Hematopor-phyrin’, Cancer, 8, 78–81 (1958).

114. R.L. Lipson, E.J. Baldes, A.M. Olsen, ‘Haematopor-phyrin Derivative: a New Aid for Endoscopic Detectionof Malignant Disease’, J. Thorac. Cardiovasc. Surg., 42,623–629 (1961).

115. R.L. Lipson, E.J. Baldes, A.M. Olsen, ‘The Use of aDerivative of Haematoporphyrin in Tumor Detection’,J. Natl. Cancer Inst., 26, 1–11 (1961).

116. R.L. Lipson, E.J. Baldes, A.M. Olsen, ‘A Further Eval-uation of the Use of Haematoporphyrin Derivative asa New Aid for Endoscopic Detection of Malignant Dis-ease’, Dis. Chest, 46, 676–679 (1964).

117. R.L. Lipson, J.H. Pratt, E.J. Baldes, M.B. Dockerty,‘Haematoporphyrin Derivative for Detection of CervicalCancer’, Obstet. Gynecol., 24, 78–84 (1964).

118. R.L. Lipson, E.J. Baldes, M.J. Gray, ‘HematoporphyrinDerivative for Detection and Management of Cancer’,Cancer, 20, 2255–2257 (1967).

119. M.J. Gray, R.L. Lipson, J.V.S. Maeck, L. Parker, D. Ro-meyn, ‘Use of Hematoporphyrin Derivative in Detec-tion and Management of Cervical Cancer: a Prelim-inary Report’, Am. J. Obstet. Gynecol., 99, 766–771(1967).

120. H.B. Gregorie, E.O. Horger, J.L. Ward, J.R. Green,T. Richards, H.C. Robertson, T.B. Stevenson, ‘Hemato-porphyrin Derivative Fluorescence in Malignant Neo-plasms’, Ann. Surg., 167, 820–828 (1968).

121. T.B. Leonard, W.L. Beck, ‘Hematoporphyrin Fluores-cence: an Aid in Diagnosis of Malignant Neoplasms’,Laryngoscopy, 81, 365–377 (1971).

122. J.F. Kelly, J.F. Snell, ‘Hematoporphyrin Derivative: aPossible Aid in the Diagnosis and Therapy of Carcinomaof the Bladder’, J. Urol., 115, 150–151 (1976).

123. M. Kreigmair, R. Baumgartner, W. Lumper, R. Reisen-berg, S. Stocker, A. Hofstetter, ‘Fluorescence Cys-toscopy Following Intravesical Instillation of 5-Amino-levulinic Acid (ALA)’, J. Urol., 149, 240A (1993).

124. P. Edholm, B. Jacobson, ‘Detection of Aortic Athero-matosis In Vivo by Reflection Spectrophotometry’, J.Atheroscler. Res., 5, 592–595 (1965).

125. C. Kitrell, R.L. Willett, C. de los Santos-Pacheco, N.B.Ratliff, J.R. Kramer, E.G. Malk, M.S. Feld, ‘Diagnosisof Fibrous Arterial Atherosclerosis Using Fluorescence’,Appl. Opt., 24, 2280–2281 (1985).

126. J. Wu, M.S. Feld, R.P. Rava, ‘Analytical Model forExtracting Intrinsic Fluorescence from a Turbid Media’,Appl. Opt., 32(19), 3585–3595 (1993).

127. C.M. Gardner, S.L. Jacques, A.J. Welch, ‘FluorescenceSpectroscopy of Tissue: Recovery of Intrinsic Fluores-cence from Measured Fluorescence’, Appl. Opt., 35(10),1780–1792 (1996).

128. H. Zeng, C. MacAulay, B. Palcic, D. McLean, ‘A Com-puterized Autofluorescence and Diffuse ReflectanceSpectroanalyzer for In Vivo Skin Studies’, Phys. Med.Biol., 38, 231–240 (1993).


129. A.F. Zuluaga, U. Utzinger, A.J. Durkin, F. Holger,A. Gillenwater, R. Jacob, B. Kemp, J. Fan, R. Richards-Kortum, ‘Fluorescence Excitation and Emission Matri-ces of Human Tissues: a System for In Vivo Measurementand Method of Data Analysis’, Appl. Spectrosc., 53(3),302–311 (1999).

130. B. Palcic, S. Lam, J. Hung et al., ‘Detection and Local-ization of Early Lung Cancer by Imaging Techniques’,Chest, 99, 742–743 (1991).

131. P.S. Anderson, S. Montan, T. Persson, S. Svanberg,S. Tapper, S.-E. Karlsson, ‘Fluorescence EndoscopyInstrumentation for Improved Tissue Characterization’,Med. Phys., 14(4), 633–636 (1987).

132. G.A. Wagnieres, A.P. Studzinski, H.E. van den Bergh,‘An Endoscopic Fluorescence Imaging System for Simul-taneous Visual Examination and Photodetection ofCancers’, Rev. Sci. Instrum., 68(1), 203–212 (1997).

133. T. Mckechnie, A. Jahan, I. Tait, A. Cuschieri, W. Sib-bett, M. Padgett, ‘An Endoscopic System for the EarlyDetection of Cancers of the Gastrointestinal Tract’, Rev.Sci. Instrum., 69(6), 2521–2523 (1998).

134. H. Zeng, A. Weiss, R. Cline, C.E. MacAulay, ‘Real-timeEndoscopic Imaging for Early Cancer Detection inthe Gastrointestinal Tract’, Bioimaging, 6(4), 151–165(1998).

135. P.S. Andersson, S. Montan, S. Svanberg, ‘MultispectralSystem for Medical Fluorescence Imaging’, IEEE J.Quant. Electron., QE-23(10), 1798–1805 (1987).

136. S. Andersson-Engels, J. Johansson, S. Svanberg, ‘Medi-cal Diagnostic System Based on Simultaneous Multispec-tral Fluorescence Imaging’, Appl. Opt., 34(1), 8022–8029(1994).

137. G.A. Duvall, J. Kost et al., ‘Laser Induced FluorescenceEndoscopy (LIFE): a Pilot Study of a Real TimeAutofluorescence Imaging System for Early Detectionof Dysplasia and Carcinoma in the GI Tract’, Endoscopy,28, S45 (1996).

138. G.A. DuVall, B.C. Wilson, N. Marcon, ‘Tissue Autoflu-orescence’, Ann. Gastrointest. Endosc., 10, 25–30 (1997).

139. R.W. Dillon, M. Goldstein, Multivariate Analysis: Meth-ods and Applications, Wiley, New York, 1984.

140. A. Albert, E.K. Harris, Multivariate Interpretation ofClinical Laboratory Data, Marcel Dekker, New York,1987.

141. R. Richards-Kortum, R.P. Rava, M. Fitzmaurice, J.R.Kramer, M.S. Feld, ‘476 nm Excited Laser InducedFluorescence Spectroscopy of Human Coronary Artery:Applications in Cardiology’, Am. Heart J., 122(4),1141–1150 (1991).

142. D.M. Haaland, ‘Multivariate Calibration Methods App-lied to Quantitative Analysis of Infrared Spectra’,in Computer-enhanced Analytical Spectroscopy, ed.P.C. Jurs, Plenum Press, New York, 1–30, Vol. 3, 1992.

143. M.T. Fahey, L. Irwig, P. Macaskill, ‘Meta-analysis of PapTest Accuracy’, Am. J. Epidemiol., 141(7), 680–689(1995).

144. M.F. Mitchell, ‘Accuracy of Colposcopy’, Consult.Obstet. Gynecol., 6(1), 70–73 (1994).

145. G. Panou-Diamandi, N.K. Uzunoglu, G. Zacharakis,G. Filippidis, T. Papazoglou, D. Koutsouris, ‘A OneLayer Tissue Fluorescence Model Based on Electro-magnetic Theory’, J. Electromag. Waves Appl., 12(8),1101–1121 (1998).

146. A.J. Durkin, R. Richards-Kortum, ‘Comparison ofMethods to Determine Chromophore Concentrationsfrom Fluorescence Spectra of Turbid Samples’, LasersSurg. Med., 19, 75–89 (1996).

147. R. Richards-Kortum, R.P. Rava, M. Fitzmaurice, L.L.Tong, N.B. Ratliff, J.R. Kramer, M.S. Feld, ‘A One-layerModel of Laser-induced Fluorescence for Diagnosis ofDisease in Human Tissue: Applications to Atheroscle-rosis’, IEEE Trans. Biomed. Eng., 36(12), 1222–1232(1989).

148. M.J.C. Van Gemert, W.M. Star, ‘Relations Between theKubelka–Munk and the Transport Equation Modelsfor Anisotropic Scattering’, Lasers Life Sci., 1, 287–298(1987).

149. H. Zeng, C. MacAulay, D.I. McLean, B. Palcic, ‘Recon-struction of In Vivo Skin Autofluorescence Spectrumfrom Microscopic Properties by Monte Carlo Simu-lations’, J. Photochem. Photobiol. B: Biol., 38(2/3),234–240 (1997).

150. J. Qu, C. MacAulay, S. Lam, ‘Laser Induced Fluores-cence Spectroscopy at Endoscopy: Tissue Optics, MonteCarlo Modeling, and In Vivo Measurements’, Opt. Eng.,34, 3334–3343 (1995).

151. S. Avrillier, E. Tinet, D. Ettori, J.M. Tualle, B. Gelebart,‘Influence of the Emission-reception Geometry on LaserInduced Fluorescence Spectra from Turbid Media’,Appl. Opt., 37(13), 2781–2787 (1998).

152. B.W. Pogue, T. Hasan, ‘Fluoropore Quantitation inTissue Simulating Media with Confocal Detection’,IEEE J. Select. Topics Quant. Electron., 12(4), 959–964(1996).

153. B.W. Pogue, G. Burke, ‘Fiber-optic Bundle Design forQuantitative Fluorescence Measurements from Tissue’,Appl. Opt., 37(31), 7429–7436 (1998).

154. J.K. Dhingra, X. Zhang, K. McMillan, S. Kabani,R. Manoharan, I. Itzkan, M.S. Feld, S.M. Shapshay,‘Diagnosis of Head and Neck Precancerous Lesions inan Animal Model Using Fluorescence Spectroscopy’,Laryngoscopy, 108(4), 471–475 (1998).

155. E.W.J. Van der Breggen, A.I. Rem, M.M. Chris-tian, C.J. Yang, K.H. Calhoun, H.J.C.M. Sterenborg,M. Motamedi, ‘Spectroscopic Detection of Oral and SkinTissue Transformation in a Model for Squamous CellCarcinoma: Autofluorescence versus AminolevulinicAcid-induced Fluorescence’, IEEE J. Select Topics, QE2,997–1007 (1997).

156. A.M. Kluftinger, N.L. Davis, N.F. Quenville, S. Lam,J. Hung, B. Palcic, ‘Detection of Squamous Cell Can-cer and Pre-cancerous Lesions by Imaging Tissue


Autofluorescence in the Hamster Cheek Pouch Model’,Surg. Oncol., 1, 183–188 (1992).

157. I. Pathak, N. Davis, Y.N. Hsiang, N.F. Quenville, B. Pal-cic, ‘Detection of Squamous Neoplasia by FluorescenceImaging Comparing Porfimer Sodium Fluorescence toTissue Autofluorescence in the Hamster Cheek PouchModel’, Am. J. Surg., 170(5), 423–426 (1995).

158. J.M. Nauta, O.C. Speelman, H.L. van Leengoed,P.G. Nikkels, J.L. Roodenburg, W.M. Star, M.J. Witjes,

A. Vermey, ‘In Vivo Photo-detection of ChemicallyInduced Pre-malignant Lesions and Squamous CellCarcinoma of the Rat Palatal Mucosa’, J. Photochem.Photobiol., B: Biol., 39(2), 156–166 (1997).

159. H.J.C.M. Sterenborg, S. Thomsen, S.L. Jacques, M. Mot-amedi, ‘In Vivo Autofluorescence of an UnpigmentedMelanoma in Mice. Correlation of Spectroscopic Prop-erties to Microscopic Structure’, Melanoma Res., 5,211–216 (1995).

Documents

Fluorescence Spectroscopy In Vivo€¦ · FLUORESCENCE SPECTROSCOPY IN VIVO 1 Fluorescence Spectroscopy In Vivo Nirmala Ramanujam University of Wisconsin, Madison, USA 1 Fluorescence