Application of Raman spectroscopy for cervical dysplasia diagnosis

FULL ARTICLE

Application of Raman spectroscopyfor cervical dysplasia diagnosis

Elizabeth M. Kanter1, Elizabeth Vargis1, Shovan Majumder2, Matthew D. Keller1,Emily Woeste3, Gautam G. Rao4, and Anita Mahadevan-Jansen*; 1

1 Department of Biomedical Engineering, Vanderbilt University, Station B, Box 351631, Nashville, TN 37235, USA2 Raja Ramanna Centre for Advanced Technology, Indore, India3 Tri-State Women’s Health Associates. 6903 Burlington Pike, Florence, KY 41042, USA4 Department of Obstetrics and Gynecology Centennial Hospital, 2300 Patterson Street, Nashville, TN 37203, USA

Received 10 September 2008, revised 6 January 2009, accepted 8 January 2009Published online 27 January 2009

Key words: optical diagnosis, hormonal status, cervix, multi-class-statistical analysis

# 2009 by WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim

1. Introduction

Cervical cancer is the second most common malig-nancy among women worldwide, with over 490 000cases diagnosed and 274 000 deaths each year [1]. Inthe US, it is estimated that 3,870 deaths will occurand 11 070 new cases of invasive cervical cancer willbe diagnosed in 2008 [2]. Due to the introduction ofearly detection methods, incidences of clinically inva-

sive cervical cancer have been shown to decrease by78% and mortality to decrease by 72% [3]. Earlyidentification of cervical precancers is the key forsuccessful treatment of the disease and for decreas-ing mortality from it.

The normal cervix is covered by two types ofepithelia: squamous and columnar. Their interface iscalled the squamo-columnar junction. Over time, thecolumnar epithelium is replaced by squamous epithe-

# 2009 by WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim

Journal of

BIOPHOTONICS

Cervical cancer is the second most common malignancyamong women worldwide, with over 490000 cases diag-nosed and 274000 deaths each year. Although currentscreening methods have dramatically reduced cervicalcancer incidence and mortality in developed countries, a“See and Treat” method would be preferred, especiallyin developing countries. Results from our previous workhave suggested that Raman spectroscopy can be used todetect cervical precancers; however, with a classificationaccuracy of 88%, it was not clinically applicable. In thispaper, we describe how incorporating a woman’s hormo-nal status, particularly the point in menstrual cycle andmenopausal state, into our previously developed classifi-cation algorithm improves the accuracy of our methodto 94%. The results of this paper bring Raman spectro-

scopy one step closer to being utilized in a clinical set-ting to diagnose cervical dysplasia.

0.0

0.5

1.0HGLGMPN

Pos

teri

orP

roba

bilit

y

Tissue Sites

0.0

0.5

1.0

HGLGMPN

Pos

teri

orP

roba

bilit

y

Tissue Sites

All Data Pre Only

0.0

0.5

1.0HGLGMPN

Pos

teri

orP

roba

bilit

y

Tissue Sites

0.0

0.5

1.0

HGLGMPN

Pos

teri

orP

roba

bilit

y

Tissue Sites

All Data Pre Only

Posterior probabilities of class membership, as deter-mined by MRDF-SMLR, for patients regardless of me-nopausal status, and for pre-menopausal patients only

* Corresponding author: e-mail: [email protected], Phone: 615-343-4787, Fax: 615-343-7919

Early View publication onwww.interscience.wiley.com(issue and page numbers not yet assigned;citable using Digital Object Identifier – DOI)

J. Biophoton. 1–10 (2009) / DOI 10.1002/jbio.200810001

lium, causing the squamo-columnar junction to movetowards the os or the opening of the ectocervix. Thistransitional epithelium is termed squamous metapla-sia [4]. Virtually all squamous cervical neoplasms be-gin at the squamo-columnar junction. The extent andlimit of their precursors coincide with the distributionof the transformation zone [5].

There are a variety of clinically normal abnormal-ities that can result in an atypical cervix. Cervicitis,or inflammation, which may be infectious or non-in-fectious, is usually the response of tissue to injuryand is a byproduct of its natural repair mechanism[5]. As described above, the transitional epitheliumor squamous metaplasia is a normal abnormality thatoccurs in the cervix. Pregnancy is another normalstate that causes a variety of biochemical and func-tional changes to occur in cervical connective tissue[6]. This process of cervical ripening results in thesoftening, dilation, and effacement of the cervix, anormal process during pregnancy that drasticallychanges the makeup of the cervix [7]. A woman’smenstrual cycle can also lead to normal changes inthe cervix. During the cycle, a woman’s hormonallevel fluctuates as levels of estrogen and progester-one peak during and after ovulation, respectively.Although these changes are small compared to thosefound during pregnancy, they do impact the bio-chemical makeup of the cervix [8]. Menopause, thepermanent cessation of the menstrual cycle, may alsoaffect the cervix due to hormonal changes. Peri-me-nopause is defined as the transitional period fromnormal menstrual periods to no periods at all, oftentaking up to 10 years.

Clinically speaking, cervical lesions can be di-vided into low grade squamous intraepithelial lesions(LGSIL) and high grade squamous intraepithelial le-sions (HGSIL). This distinction is important, as pa-tients with low grade lesions are usually followedbut not treated, while patients with high grade le-sions are usually treated immediately with extendedfollow-up. Patients with specific strains of human pa-pillomavirus (HPV) are typically placed in the samecategory as patients with low grade lesions and aretreated as such. Certain strains of HPV may be in-volved in the early stages of cervical cancer, whileother strains may aid in the progression of the dis-ease [9]. These viruses infect skin and mucosal mem-branes to produce characteristic epithelial prolifera-tion, which can have the capacity to undergomalignant transformations [5].

Currently, the primary screening tool for cervicalprecancer is the Papanicolaou (Pap) smear, wherecells scraped from the walls of the ecto- and endo-cervix are examined and diagnosed [10]. The wide-spread application of the Pap smear as a screeningtool has greatly decreased the incidence of cervicalcancer, especially in the US [11]. While the specifi-city of the Pap smear is generally very high (95%)

[11], its sensitivity can be as low as 20% to 50% de-pending on the prevalence of the disease within thepopulation [12]. Colposcopy usually follows an ab-normal Pap smear in the US and is used to directthe taking of biopsies [13]. After applying 4–6%acetic acid to the cervix to turn abnormal areaswhite, a colposcope is used to visualize these abnor-mal areas on the surface of the cervix. Multiple biop-sies are then taken using standard punch biopsy for-ceps, and these tissue samples are fixed in formalinand sent for histological examination. Colposcopyhas a sensitivity of 62% to 98% in pin-pointing thepresence and grade of lesions [14–16]. The result ofthe biopsy decides treatment of the disease. Thisprocedure requires extensive training, is time-con-suming, and its sensitivity is variable and limited,even in the hands of expert practitioners [17]. A“see and treat” tool that can provide high sensitivityin diagnosing cervical cancer would be optimal inaddressing the limitations of colposcopy-guidedbiopsy.

Optical methods are prime candidates for “seeand treat” procedures for the cervix because theycan provide automated, fast, and non-intrusive char-acterization of normal and abnormal tissues in vivo.Choosing the right optical method depends on theparticular requirements of a given problem. For ex-ample, optical coherence tomography (OCT) canproduce high-resolution, cross-sectional images, butsince its ability to detect disease depends on visualiz-ing changes in the tissue microstructure, it is difficultfor OCT to detect early cases of cervical precancer[18]. Fluorescence spectroscopy has been studied ex-tensively for screening and diagnosing cervical pre-cancers [19–23], but the results have had unaccepta-bly high false positive rates for detecting benignabnormalities such as inflammation, hyperplasia, andmetaplasia [24, 25]. Previous studies using infraredmicrospectroscopy have shown differences in spectradue to the normal, cyclical changes of cervicalepithelium [26]. However, this application is limitedbecause tissue must be removed before spectra areacquired. For a tissue like the cervix, where manynormal changes occur, the optical technique appliedmust be able to discern among the various condi-tions such as cervicitis, LGSIL, and HGSIL. Ramanspectroscopy is a molecular specific technique pro-viding information about the biochemical composi-tion of a tissue by probing vibrational and rotationalbond transitions in various biomolecules [27]. Ourprevious work suggests that Raman spectroscopycombined with automated multivariate statisticalanalysis can be used to detect changes in tissue asso-ciated with the progression of cervical cancer [28].With a sensitivity of 88%, our previous applicationcannot be applied within the clinic since it does notprovide sufficient improvement over traditionalmethods. While this may be due to Raman spectro-

E. M. Kanter et al.: Application of Raman spectroscopy for cervical dysplasia diagnosis2

Journal of

BIOPHOTONICS

# 2009 by WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim www.biophotonics-journal.org

scopy’s inability to be utilized in this capacity, we be-lieve that including small, normal variations like me-nopausal or hormonal status in our classification al-gorithms will improve the sensitivity to over 90%.

The goal of this paper is to examine the effect ofincorporating normal variations into our classifica-tion algorithm prior to data analysis. Raman spectrawere acquired from 145 patients undergoing either aroutine Pap smear or biopsy guided by colposcopy.The effects that hormonal status, particularly thepoint in the menstrual cycle and menopausal state,had on Raman spectra gathered from normal cervixareas were examined. These effects were then ac-counted for to improve the classification ability of apreviously developed statistical algorithm for detect-ing cervical precancer, highlighting the need to ac-count for such variation in any further analysis.

2. Materials and Methods

2.1. Clinical Study Design-Pap SmearPatients

A total of 102 patients undergoing a routine Papsmear were recruited to participate in the study asapproved by the Copernicus and Vanderbilt Institu-tional Review Boards (IRBs). To be eligible for en-rollment, the patient had to be undergoing a routinePap smear, be between the ages of 18–75, and stillhave a cervix. Informed consent was obtained fromeach patient prior to the procedure. The cervix wasexposed and visually examined by the attending phy-sician, and the Pap procedure was done according tostandard clinical protocol. The cervix was wipedclean with a dry cotton swab, after which Ramanmeasurements were taken from three locations onthe ectocervix. The spectra were considered normalif the Pap smear was negative. The patient’s age,date of last menstrual period, use of artificial hor-mones, menopausal status, and any previous abnor-mal Pap smears were all noted upon chart review.

2.2. Clinical Study Design-DysplasiaPatients

A total of 43 patients undergoing colposcopy-guidedbiopsy were recruited to participate in the study asapproved by the Copernicus and Vanderbilt IRBs.To be eligible for enrollment, the patient had to beundergoing a colposcopy-guided biopsy, be betweenthe ages of 18–75, and still have a cervix. After in-formed consent was obtained from each patient, thecervix was exposed and visually examined by the at-

tending physician. Acetic acid was applied to the cer-vix to turn abnormal areas white for visualization,and spectra were acquired from multiple areas of ab-normal tissue and 1–2 visually normal areas. Aceto-white tissue was then removed and placed in fixativesolution for pathology examination. Based onpathology, spectra were placed into four categoriesfor analysis: normal, metaplasia, LGSIL, andHGSIL.

2.3. Instrumentation

Raman spectra were collected from multiple sites invivo using a portable Raman spectroscopy systemconsisting of a 785 nm diode laser (Process Instru-ments, Inc., Salt Lake City, UT), a beam-steered fi-ber optic probe (Visionex Inc., Atlanta, GA), animaging spectrograph (Kaiser Optical Systems, Inc.,Ann Arbor, MI), and a back-illuminated, deep-de-pletion, thermo-electrically cooled charge coupleddevice (CCD) camera (Roper Scientific, Inc., Prince-ton, NJ), all controlled with a laptop computer. De-tails of the system have been previously reported[30]. For each measurement, the fiber optic probedelivered 80mW of incident light onto the tissue for3 seconds, and the overhead fluorescent lights wereturned off. Spectra were obtained from the surfaceof the cervix. As the penetration depth is about300 mm, information from both the epithelium andthe stroma are present in the spectra.

Spectral calibration of the system was performedeach day using a neon-argon lamp and naphthaleneand acetaminophen standards to correct for systemwavenumber, laser excitation, and throughput varia-tions. The spectra were processed for fluorescencesubtraction and noise smoothing using the modifiedpolynomial fit method described previously [29]. Fol-lowing data processing, each spectrum was normal-ized to its mean spectral intensity across all Ramanbands to account for overall intensity variability.These normalized spectra were categorized accord-ing to menopausal status and histopathological clas-sification as determined by the pathologist.

2.4. Statistical Analysis

The two-step multivariate analysis technique used inthis paper has been previously described [30]. In thefirst step, using maximum representation and discri-mination feature (MRDF), the processed data setundergoes a two-part, non-linear transform to ex-tract relevant features that provide the best class se-paration. The second step uses sparse multinomiallogistic regression (SMLR) for classifying the MRDF

J. Biophoton. (2009) 3

FULLFULLARTICLEARTICLE

# 2009 by WILEY-VCH Verlag GmbH & Co. KGaA, Weinheimwww.biophotonics-journal.org

output features into corresponding tissue categories.SMLR is a probabilistic multi-class model based ona Bayesian machine-learning framework of statisticalpattern recognition, which separates a set of labeledinput data into their classes by predicting the poster-ior probabilities of class membership. An unbiasedclassification algorithm was developed using thesetwo methods and was tested using leave-one-patient-out cross-validation. The classification accuracy, de-fined as the total number of spectra correctly classi-fied divided by the total number of spectra, is usedto determine the validity of the method. The use ofacetic acid should not affect the analysis because sta-tistical comparisons were done with either the pre-sence or absence of acetic acid unless otherwisenoted.

3. Results

3.1.Variations in Menopausal Status

The Raman spectra in Figure 1 are shown stratifiedby menopausal status into four groups: pre-menopau-sal proliferative phase (days 1–14 of the menstrualcycle) or pre-menopausal before ovulation (PBO);pre-menopausal secretory phase (days 15–28+ of themenstrual cycle) or pre-menopausal after ovulation(PAO); peri-menopausal (PERI); and post-menopau-sal (POST). These terms are defined in Table 1. Somespectra were thrown out due to slipping of the probeor if mucus was present on the cervix. From the spec-tra that remained, there were subtle but consistent

differences according to location within the menstrualcycle and menopausal status, especially at the1250 cm�1 peak, typically associated with collagen,and the 1300–1320 cm�1 region, typically associatedwith cellular features [31]. As seen in the confusionmatrix in Table 2, the spectra were classified intotheir respective hormonal and menopausal groups bythe MRDF-SMLR algorithm with an overall accuracyof 98%. The number along the diagonal of the confu-sion matrix represents the number of spectra thatwere classified correctly. In this case, only three spec-tra were misclassified.

3.2. Dysplasia Study

Average spectra from cervical tissues categorized asLGSIL, HGSIL, normal, and metaplasia are dis-played in Figure 2. These spectra were obtained be-fore colposcopy-guided biopsy. The most prominentspectral differences among the classes are seen inthe region between 1230 cm�1 and 1300 cm�1, wherea number of biological molecules contribute to thespectral features. Using the MRDF-SMLR algo-rithm, the overall classification accuracy was 88%,with a sensitivity of 86% and specificity of 97% forseparating diseased from non-diseased tissues [32].These results are somewhat undesirable for use inthe clinic, especially in the LGSIL category, wherethe misclassification rate was 19%. Based on the re-sults of the above study, the spectra were further

1000 1100 1200 1300 1400 1500 1600 17000

0.5

1

1.5

2

2.5

3

3.5

Wavenumber (cm-1)

Intens

ity(au)

PBO

PAOPERI

POST

Figure 1 Average Raman spectra for post menopausal nor-mal cervix (POST-29), peri menopausal normal cervix(PERI-34), pre-menopausal after ovulation normal cervix(PAO-54) and pre-menopausal before ovulation normalcervix (PBO-47).

Table 1 Description of abbreviations for menopausal andhormonal status.

Category Abbreviation Age Time sincelast period

Pre-menopausalbefore ovulation

PBO 18–45 5–14 days

Pre-menopausalafter ovulation

PAO 18–45 15–28 days

Peri-menopausal PERI 45–55 Variableperiods

Post-menopausal POST 50þ 2þ years

Table 2 Classification of different hormonal variationsusing MRDF and SMLR leave-one- patient-out cross-vali-dation.

Histological Classification

PBO PAO PERI POST

RamanClassification

PBO 47 0 0 0PAO 0 53 0 0PERI 0 0 33 1POST 0 1 1 28


Journal of

BIOPHOTONICS


stratified by menopausal status with only pre-meno-pausal spectra used. Table 3 shows the confusion ma-trix comparing Raman diagnosis to histopathologydiagnosis using the following four classes: normal,metaplasia, LGSIL, and HGSIL. In this analysis,only six out of 95 spectra were misclassified – twoLGSIL and four normal. This stratification of spec-tra based on menopausal state results in an improve-ment of overall classification accuracy from 88% to94%. This result is shown in Figure 3, which displaysthe posterior probabilities of class membership, asdetermined by MRDF-SMLR, for all patients re-gardless of menopausal status (as published by Kan-ter et al. [32]), and for pre-menopausal patientsonly. Besides the improvement in classification rate,Figure 3 also shows an increase in the algorithm’sconfidence in its classification, as more spectra re-ceive probabilities of class membership closer toone.

3.3. Independent Validation

Based on the results of the previous study, a smallindependent validation was performed. Spectra ob-tained from patients undergoing routine Pap smearthat had resulted in an abnormal diagnosis wereclassified using the above algorithm (from the dys-plasia study). When the spectra were taken, the loca-tion of the measurement on the cervix was noted.The spectra were then run through the MRDF-SMLR algorithm developed for the dysplasia classifi-cation using only the pre-menopausal data. A colpo-scopy was done as normal, and those results werecompared to the algorithm results, as shown in Ta-ble 4. Of the six spectra that were analyzed in thismanner, three were classified as LGSIL and three asnormal. All of the spectra came from a cervix thatthe Pap smear suggested may contain areas ofLGSIL. In all areas that were classified by the algo-rithm as normal, no biopsies were taken, and thecolposcopic examination suggested that these areaswere indeed normal. In the areas that classified as

1000 1100 1200 1300 1400 1500 1600 17000

0.5

1

1.5

2

2.5

3

Wavenumber (cm-1)

Intens

ity(au)

HGSIL

LGSILMetaplasia

Normal

Figure 2 Average Raman spectra for normal ectocervix,metaplasia, LGSIL and HGSIL.

Table 3 Classification of pre-menopausal samples- normal,metaplasia, LGSIL and HGSIL using MRDF and SMLRleave-one-patient-out cross-validation.

Histological Classification

Normal Meta-plasia

LGSIL HGSIL

RamanClassifi-cation

Normal 27 0 2 0Metaplasia 0 12 0 0LGSIL 4 0 39 0HGSIL 0 0 0 11

0.0

0.5

1.0HGLGMPN

Pos

teri

orP

roba

bilit

y

Tissue Sites

0.0

0.5

1.0

HGLGMPN

Pos

teri

orP

roba

bilit

y

Tissue Sites

All Data Pre Only

0.0

0.5

1.0HGLGMPN

Pos

teri

orP

roba

bilit

y

Tissue Sites

0.0

0.5

1.0

HGLGMPN

Pos

teri

orP

roba

bilit

y

Tissue Sites

All Data Pre Only

Table 4 Results of the independent validation.

Raman Classification Pathology

LGSIL LGSIL confirmed by biopsyLGSIL LGSIL confirmed by biopsyLGSIL UnknownNormal Suspected NormalNormal Suspected NormalNormal Suspected Normal

Figure 3 Posterior probabilities ofclassification as normal ectocervix(N), metaplasia (MP), LGSIL (LG)and HGSIL (HG) from the entiredata set (classification accuracy88%) and the pre-menopausal dataonly (classification accuracy 94%)[33].




LGSIL, two of these regions were biopsied and con-firmed to be LGSIL. The other region was not biop-sied, but was noted in the examination as having amosaic appearance, meaning it may have had somedisease, although there is no way to confirm this.

4. Discussion

Raman spectroscopy can be used in the cervix, de-spite variations in both patient history and physiol-ogy. This study brings Raman spectroscopy onestep closer to clinical use by improving its specifi-city in diagnosing cervical dysplasia. This improve-ment was accomplished by incorporating variationsin the normal cervix to differentiate LGSIL,HGSIL, and metaplasia from normal tissues. Ourresults suggest that Raman spectroscopy is sensitiveenough to detect subtle changes in the cervix suchas those that occur due to hormonal or menopausalstatus as well as small changes that are associatedwith dysplasia.

Previous fluorescence studies have shown thatthere is a statistical difference between pre and post-menopausal fluorescence data and that the post-me-nopausal women have a higher average fluorescencesignal [33] which may need to be considered whenusing fluorescence spectroscopy for disease detec-tion. Another study suggests that fluorescence spec-troscopy is predominantly affected by hemoglobinabsorption and these effects can be avoided if meas-urements are not taken during the first eight daysof the cycle [34], indicating that fluorescence spec-troscopy is not sensitive to subtle, hormonal changesin the cervix. Fluorescence spectroscopy is capableof distinguishing HGSIL from normal, however inone study, the sensitivity and specificity of detectingsquamous normal cervix tissue and LGSIL of thecervix in 161 patients are only 55% and 63%, re-spectively [35] and thus fluorescence spectroscopycannot be used to find early indicators of disease.Raman spectroscopy at first glance seems to have si-milar results to fluorescence spectroscopy. But sinceRaman is very sensitive to biochemical changes, itshould be able to discriminate small changes asso-ciated with disease and small changes associatedwith other factors. Therefore, it is necessary to ac-count for such changes, such as menopausal status,when using Raman to detect early disease whereonly a few disease-related biochemical changes haveoccurred.

This new approach was necessary because hor-monal changes, such as menopausal status and loca-tion in the menstrual cycle, have the potential tochange the composition of the ectocervix [36]. Theectocervix consists primarily of collagenous connec-tive tissue, which is approximately 15% smooth

muscle, a small amount of elastic tissue, and aground substance of mucopolysaccharides. Duringthe menstrual cycle, the cervix becomes softer andmore elastic as the level of estrogen increases. Afterovulation, this process is reversed and the cervixloses some of its elasticity. During peri-menopause,the layer of epithelial cells thins and the vascularityand cellular content of the cervix is erratic, but thespectra remain consistent. The most variable andtherefore the hardest group to classify is the post-menopausal group. The absence of ovarian estrogenand progesterone causes the cervix to change, lead-ing to both dryness and atrophy, although theseconditions are considered normal in a woman whohas gone through menopause. The cervical Ramansignatures vary significantly depending on locationwithin the menstrual cycle and with the onset andcompletion of menopause. These spectral differ-ences are shown in Figure 1; the most notable dif-ferences occur around 1250 cm�1, 1300 cm�1, and1320 cm�1, most likely due to changes in proteinlevels, especially elastin and collagen, for reasonsnoted above. The 98% classification accuracy pro-vided by the MRDF-SMLR algorithm for discrimi-nating spectra into PBO, PAO, PERI, and POST ca-tegories indicates the validity and necessity ofincluding hormonal variations when analyzing cervi-cal Raman spectra.

As shown in Figure 2, changes in Raman spectradue to cervical dysplasia are different than those as-sociated with hormonal variations. The largest differ-ence when comparing LGSIL Raman spectra to nor-mal Raman spectra is in the 1230–1300 cm�1 region.This peak range is usually associated with proteins(amide III), DNA (guanine), and lipids (CH trans-formations) [31]. It is expected that there will be var-iations in the protein and lipid content when dyspla-sia occurs, as well as an increase in the ratio ofnucleus to cytoplasm due to increased DNA contentin dysplastic cells. Another expected change is a re-duction in glycogen peaks that occur around1300 cm�1 [37]. This difference is expected to beminimal in LGSIL because the disease affects only asmall portion of the epithelium. As the disease pro-gresses towards HGSIL, this drop in the glycogenpeak is expected to become more drastic.

As indicated in Figure 3, restricting the dysplasiadiagnostic algorithm to only pre-menopausal spectraresulted in improving the overall accuracy for discri-minating among normal, metaplasia, LGSIL, andHGSIL from 88% to 94%. Although there were in-sufficient numbers of PERI and POST spectra toachieve statistical significance, it is expected thatsuch analyses would achieve similar results. Due tothe large effect of hormonal and menopausal statuson the spectra, the optimal strategy for future clini-cal implementation would first stratify by hormonal/menopausal status before applying a separate diag-


Journal of

BIOPHOTONICS


nostic algorithm for each hormonal/menopausal ca-tegory.

The classification technique used in this paper, anon-linear multi-class algorithm, yields a posteriorprobability of how likely the spectra are to fit into aparticular tissue category. This powerful tool quanti-fies the confidence with which a sample is classifiedcorrectly with Raman spectroscopy. If the value ishigh, a biopsy may not be needed; if the probabilityis low and the area is suspicious based on the doc-tor’s observation, a biopsy could be taken as is thecurrent clinical protocol. Results from the small in-dependent validation set provide good examples ofthese ideas although this study has several limita-tions (biopsies cannot be taken from normal subjectsand the spot the measurement is taken from is ap-proximate). In patients with abnormal Pap smears,Raman measurements were taken from several areasof the cervix, and the previously developed diagnos-tic algorithm successfully classified the spectra in ac-cordance with colposcopy and/or biopsy results.Since these patients had LGSIL, the biopsies wereultimately unnecessary and could have been avoidedbased on their Raman diagnoses.

This application of Raman spectroscopy wouldbe particularly useful in developing countries, where“see and treat” methods are optimal. One majorproblem with screening alone is poor follow-up test-ing among women with abnormal Pap smears.Usually, an abnormal Pap smear requires a follow-up biopsy and a return visit 3–6 months later de-pending on the result, but an estimated 10–61% ofwomen with abnormal Pap smears do not show upfor follow-up testing [38]. Additionally, only an esti-mated 19% of women in developing countries havebeen screened for cervical dysplasia in the past fiveyears, compared with around 60% of women in de-veloped countries [39]. In developing countries, stand-ard practice is for a nurse to photograph the cervixto send to a doctor for diagnosis, but this time-con-suming process also has a high error rate and levelof ambiguity. Using Raman spectroscopy wouldallow a nurse to find suspicious areas, take Ramanmeasurements, report an accurate diagnosis, and de-cide on appropriate treatment all on the same day.This process could reduce the number of patientswho are treated unnecessarily and ensure that allpatients receive the appropriate treatment. Besidesdeveloping countries, this technique for cervical dys-plasia diagnosis would greatly benefit rural commu-nities and has to the potential to reduce the numberof cervical biopsies taken.

Acknowledgements The authors would like to acknowl-edge the financial support of the NCI/NIH (R01-CA95405). We would also like to thank the doctors andstaff at Vanderbilt University Medical Center and Tri-stateWomen’s Health for all their help.

Elizabeth Kanter receivedher B.E. in biomedical engi-neering from VanderbiltUniversity in 2003, her M.S.from the University of Ari-zona in 2005 and her Ph.D.from Vanderbilt Universityin 2008. Her Ph.D. researchfocused on detection of cer-vical pre-cancers using Ra-man Spectroscopy. She cur-rently holds a post-doctoralposition at the University ofTexas-Dallas.

Elizabeth Vargis receivedher B.S. in bioengineeringfrom the University of Cali-fornia at Berkeley in 2004and her M.S. in biomedicalengineering from VanderbiltUniversity in 2007. She alsoworked at the LawrenceBerkeley National Lab from2003–2005 where she fo-cused on understanding therelationship between levelsof hormone receptors and

the occurrence of breast cancer. Elizabeth is currentlypursuing a Ph.D. in biomedical engineering from Vander-bilt looking at how normal changes in the cervix affectthe ability of Raman spectroscopy to detect cervical ab-normalities and precancers.

Shovan K. Majumder re-ceived his B.Sc (with Hon-ors) and M.Sc degrees inPhysics from Jadavpur Uni-versity, Kolkata, India. Aftergraduating (post M.Sc) fromBhabha Atomic ResearchCentre (BARC) TrainingSchool, Mumbai, India in1992, he joined Raja Ra-

manna Centre for Advanced Technology, Indore, aDept. of Atomic Energy Laboratory of Govt. of India.While working there in the Laser Biomedical Applica-tions and Instrumentation Division as Scientist sincethen, he received his Ph.D. degree from Devi AhilyaUniversity, Indore, India for his work on “Laser In-duced Fluorescence Spectroscopy for Cancer Diagno-sis” and held a postdoctoral position for two years atthe Biomedical Optics Laboratory of Vanderbilt Uni-versity. His research interest includes applications offluorescence, diffuse reflectance and Raman spectro-




References

[1] G. P. Parham, V. V. Sahasrabuddhe, M. H. Mwanaha-muntu, B. E. Shepherd, M. L. Hicks, E. M. Stringer,and S. H. Vermund, Prevalence and predictors ofsquamous intraepithelial lesions of the cervix in HIV-infected women in Lusaka, Zambia. Gynecologic On-cology 103, 1017–1022 (2006).

[2] Cervical Cancer Resource Center (2007).[3] G. H. Anderson, D. A. Boyes, J. L. Benedet, J. C. Le

Riche, J. P. Matisic, K. C. Suen, A. J. Worth, A. Mill-ner, and O. M. Bennett, Organisation and results ofthe cervical cytology screening programme in BritishColumbia, 1955–1985. Br. Med. J. (Clin. Res. Ed.).296, 975–978 (1988).

[4] K. Krantz, The Anatomy of the Human Cervix, Grossand Microscopic (University of Chicago Press, Chica-go 1973).

[5] T. Wright, R. Kurman, and A. Ferenczy, Cervical In-traepithelial Neoplasia (Springer-Verlag, New York,1994).

[6] P. C. Leppert, Anatomy and physiology of cervical ri-pening. Clin. Obstet. Gynecol. 38, 267–279 (1995).

[7] H. Maul, G. Saade, and R. E. Garfield, Prediction ofterm and preterm parturition and treatment monitor-ing by measurement of cervical cross-linked collagenusing light-induced fluorescence. Acta Obstet. Gyne-col. Scand. 84, 534–536 (2005).

[8] D. A. Eschenbach, S. S. Thwin, D. L. Patton, T. M.Hooton, A. E. Stapleton, K. Agnew, C. Winter, A.Meier, and W. E. Stamm, Influence of the normalmenstrual cycle on vaginal tissue, discharge, and mi-croflora. Clin. Infect. Dis. 30, 901–907 (2000).

[9] D. V. Coleman, C. Wickenden, and A. D. Malcolm,Association of human papillomavirus with squamouscarcinoma of the uterine cervix. Ciba Found. Symp.120, 175–189 (1986).

scopy for biomedical optical diagnosis and imaging.Other interests include applications of pattern recogni-tion based methods for evaluation and development ofdiscrimination algorithms for classification of tissuepathologies.

Matt Keller received hisB.E. and M.S. degrees inbiomedical engineering fromVanderbilt in 2003 and 2006,respectively. He is currentlya Ph.D. candidate in thesame field at Vanderbilt.Matt has been supported inhis graduate work by a Ho-ward Hughes Medical Insti-tute Pre-doctoral Fellowshipand Department of Defense

Breast Cancer Research Program Pre-doctoral Fellow-ship. His current research interest is the use of Ramanspectroscopy for evaluating margin status during breastconserving surgery.

Gautam G. Rao received hisB.S. and M.D. degrees fromthe University of Miami. Hecompleted his residency inObstetrics and Gynecologyat Wayne State Universityin Detroit, Michigan and fel-lowship in Gynecologic On-cology at the University ofTexas Southwestern MedicalCenter in Dallas, Texas. Heserved on the faculty in the

Department of Obstetrics and Gynecology at Vander-bilt University Medical Center from 2005–2007. He isnow in practice at The Sarah Cannon Cancer Center atCentennial Medical Center in Nashville, TN.

Emily Woeste is a Board Cer-tified Obstetrician and Gyne-cologist in a private practicein the Northern Kentuckycommunity. She has beenwith her practice for twoyears. She completed resi-dency at The Ohio State Uni-versity in 2006. She attendedmedical school at the Univer-sity of Kentucky College ofMedicine and graduated in

2002. She graduated with a Bachelor in Science from theUniversity of Kentucky in 1998. In her spare time, she en-joys spending time with her family, reading and running.

Anita Mahadevan-Jansenreceived her B.S. and M.S.degrees in physics from theUniversity of Bombay,Bombay, India, in 1988 and1990, respectively. She re-ceived her M.S. and Ph.D.degrees in biomedical engi-neering from the Universityof Texas at Austin in 1993and 1996, respectively. Shejoined the faculty of theDepartment of Biomedical

Engineering at Vanderbilt University, in the fall of1998. Her expertise is in the area of optical spectro-scopy and imaging, specifically on the application offluorescence and Raman spectroscopy for the detec-tion of tissue physiology, as well as pathologies such ascancers.


Journal of

BIOPHOTONICS


[10] B. Bates, A Guide to Physical Examination (J. B. Lip-pincott Co., Philadelphia, 1974).

[11] E. R. Myers, D. C. McCrory, S. Subramanian,N. McCall, K. Nanda, S. Datta, and D. B. Matchar, Set-ting the target for a better cervical screening test: char-acteristics of a cost-effective test for cervical neoplasiascreening. Obstet. Gynecol. 96, 645–652 (2000).

[12] M. T. Fahey, L. Irwig, and P. Macaskill, Meta-analysisof Pap test accuracy. Am J. Epidemiol. 141, 680–689(1995).

[13] L. Burke, D. A. Antonioli, and B. S. Ducatman, Col-poscopy, text and atlas. In. (Appleton and Large, Nor-walk, 1991).

[14] U. Minoru, Cervical Adenocarcinoma: A ColposcopicAtlas (Ishiyaku – EuroAmerica Inc., St. Louis, 1985).

[15] O. Kierkegaard, C. Byrjalsen, K. H. Frandsen, K. C.Hansen, and M. Frydenberg, Diagnostic accuracy ofcytology and colposcopy in cervical squamous intra-epithelial lesions. Acta Obstetricia et GynecologicaScandinavica 73, 648–651 (1994).

[16] J. Monsonego, J. Pintos, C. Semaille, M. Beumont,R. Dachez, L. Zerat, A. Bianchi, and E. Franco, Hu-man papillomavirus testing improves the accuracy ofcolposcopy in detection of cervical intraepithelial neo-plasia. International Journal of Gynecological Cancer.16, 591–598 (2006).

[17] M. Mitchell, Accuracy of Colposcopy. Consultationsin Obstetrics and Gynecology 6 (1994).

[18] P. F. Escobar, L. Rojas-Espaillat, S. Tisci, C. Enerson,J. Brainard, J. Smith, N. J. Tresser, F. I. Feldchtein,L. B. Rojas, and J. L. Belinson, Optical coherence to-mography as a diagnostic aid to visual inspection andcolposcopy for preinvasive and invasive cancer of theuterine cervix. International Journal of GynecologicalCancer. 16, 1815–1822 (2006).

[19] S. K. Chang, Y. N. Mirabal, E. N. Atkinson, D. Cox,A. Malpica, M. Follen, and R. Richards-Kortum, Com-bined reflectance and fluorescence spectroscopy for invivo detection of cervical pre-cancer. J. Biomed Opt.10, 024031 (2005).

[20] S. K. Chang, I. Pavlova, N. M. Marin, M. Follen, andR. Richards-Kortum, Fluorescence spectroscopy as adiagnostic tool for detecting cervical pre-cancer. Gy-necol. Oncol. 99, S61–63 (2005).

[21] U. Utzinger, E. V. Trujillo, E. N. Atkinson, M. F. Mit-chell, S. B. Cantor, and R. Richards-Kortum, Perform-ance estimation of diagnostic tests for cervical pre-cancer based on fluorescence spectroscopy: effects oftissue type, sample size, population, and signal-to-noise ratio. IEEE Trans. Biomed. Eng. 46, 1293–1303(1999).

[22] J. S. Lee, O. Shuhatovich, R. Price, B. Pikkula, M. Fol-len, N. McKinnon, C. Macaulay, B. Knight, R. Ri-chards-Kortum, and D. D. Cox, Design and prelimin-ary analysis of a study to assess intra-device andinter-device variability of fluorescence spectroscopyinstruments for detecting cervical neoplasia. Gynecol.Oncol. 99, S98–111 (2005).

[23] M. F. Mitchell, S. B. Cantor, N. Ramanujam, Tortolero-Luna G, Richards-Kortum R. Fluorescence spectro-

scopy for diagnosis of squamous intraepithelial lesionsof the cervix. Obstet. Gynecol. 93, 462–470 (1999).

[24] N. Ramanujam, M. F. Mitchell, A. Mahadevan-Jansen,S. L. Thomsen, G. Staerkel, A. Malpica, T. Wright,N. Atkinson, and R. Richards-Kortum, Cervical pre-cancer detection using a multivariate statistical algo-rithm based on laser-induced fluorescence spectra atmultiple excitation wavelengths. Photochem. Photo-biol. 64, 720–735 (1996).

[25] K. T. Schomacker, J. K. Frisoli, C. C. Compton, T. J.Flotte, J. M. Richter, N. S. Nishioka, and T. F. Deutsch,Ultraviolet laser-induced fluorescence of colonic tissue:basic biology and diagnostic potential. Lasers. Surg.Med. 12, 63–78 (1992).

[26] M. J. Romeo, B. R. Wood, and D. McNaughton, Ob-serving the cyclical changes in cervical epitheliumusing infrared microspectroscopy. Vibrational Spec-troscopy 28, 167 (2002).

[27] R. J. Colthrup. Infrared and Raman spectroscopy. In.(1991).

[28] M. Keller, E. Kanter, and A. Mahadevan-Jansen, Ra-man spectroscopy for cancer diagnosis. Spectroscopy21(9), 33 (2006).

[29] C. A. Lieber and A. Mahadevan-Jansen, Automatedmethod for subtraction of fluorescence from biologi-cal Raman spectra. Applied Spectroscopy 57, 1363–1367 (2003).

[30] S. K. Majumder, S. Gebhart, M. D. Johnson,R. Thompson, W. C. Lin, and A. Mahadevan-Jansen,A probability-based spectroscopic diagnostic algo-rithm for simultaneous discrimination of brain tumorand tumor margins from normal brain tissue. Appl.Spectrosc. 61, 548–557 (2007).

[31] R. A. Bitar, S. Had Martinho, C. J. Tierra-Criollo, L. N.Zambelli Ramalho, M. M. Netto, and A. A. Martin,Biochemical analysis of human breast tissues usingFourier-transform Raman spectroscopy. J. Biomed.Opt. 11, 054001 (2006).

[32] E. M. Kanter, S. Majumder, E. Vargis, A. Robichaux-Viehoever, G. J. Kanter, and H. Shappell, III HWJ,Mahadevan-Jansen A. Multi-class discrimination ofcervical precancers using Raman spectroscopy. Jour-nal of Raman Spectroscopy (In Press) 2008.

[33] E. M. Gill, A. Malpica, R. E. Alford, A. R. Nath,M. Follen, R. R. Richards-Kortum, and N. Ramanu-jam, Relationship between collagen autofluorescenceof the human cervix and menopausal status. Photo-chemistry and Photobiology 77, 653–658 (2003).

[34] D. D. Cox, S. K. Chang, M. Y. Dawood, G. Staerkel,U. Utzinger, R. R. Richards-Kortum, and M. Follen,Detecting the signal of the menstrual cycle in fluores-cence spectroscopy of the cervix. Applied Spectro-scopy 57, 67–72 (2003).

[35] S. K. Chang, I. Pavlova, N. M. Marin, M. Follen, andR. Richards-Kortum, Fluorescence spectroscopy as adiagnostic tool for detecting cervical pre-cancer. Gy-necologic Oncology 99, S61–S63 (2005).

[36] V. L. Katz, R. A. Lobo, G. Lentz, and D. Gershenson,Comprehensive Gynecology, 5th ed. (Elsevier Inc.,Katz, 2008).




[37] J. Sellors and R. Sankaranarayanan, Colpsocopy andTreatment of Cervical Intreaepithelial Neoplasia: ABeginners’ Manual. In. (International Agency for Re-search on Cancer, Lyon, 2003).

[38] E. Shinn, K. Basen-Engquist, T. Le, A. Hansis-Diarte,D. Bostic, J. Martinez-Cross, A. Santos, and M. Follen,Distress after an abnormal Pap smear result: scale de-

velopment and psychometric validation. Prev. Med.39, 404–412 (2004).

[39] E. Gakidou, S. Nordhagen, and Z. Obermeyer, Cover-age of Cervical Cancer Screening in 57 Countries:Low Average Levels and Large Inequalities. PLoSMedicine 5, e132 (2008).


Journal of

BIOPHOTONICS


Documents

Application of Raman spectroscopy for cervical dysplasia diagnosis