Analysis of Subtraction Methods Contrast …...Abstract Analysis of Subtraction Methods in 3D Contrast-Enhanced MR Digital Subtraction Angiography Yuexi Huang Master of Science 2001

Analysis of Subtraction Methods in 3D Contrast-Enhanced

MR Digital Subtraction Angiography

Yuexi Huang

A thesis submitted in conformity with the requirements for the degree of Master of Science, Department of Medical Biophysics,

University of Toronto

O Copyright by Yuexi Huang 2001

National Library of Canada

Bibliothèque nationaIe du Canada

Acquisitions and Acquisitions et Bibliographie Services services bibliographiques 395 Wellington Street 395. rue Wellington Ottawa O N K1A ON4 OttawaON KlAON4 Canada Canada

The author has granted a non- exclusive licence allowing the National Library of Canada to reproduce, loan, distn'bute or sell copies of this thesis in microform, paper or electronic formats.

The author retains ownership of the copyright in this thesis. Neither the thesis nor substantial extracts fiom it may be printed or otherwise reproduced without the author's permission.

L'auteur a accordé une licence non exclusive permettant à la Bibliothèque nationale du Canada de reproduire, prêter, distribuer ou vendre des copies de cette thèse sous la forme de microfiche/fïlm, de reproduction sur papier ou sur format électronique.

L'auteur conserve la propriété du droit d'auteur qui protège cette thèse. Ni la thèse ni des extraits substantiels de celle-ci ne doivent être imprimés ou autrement reproduits sans son autorisation.

Abstract

Analysis of Subtraction Methods in 3D Contrast-Enhanced MR Digital Subtraction Angiography

Yuexi Huang Master of Science 2001

Department of Medical Biophysics University of Toronto

3D Contrast-enhanced MR angiography is promising as a replacement to

catheter-based x-ray angiography due to its non-invasive nature. However, the

image quality of MR angiography is still inferior to that of x-ray angiography at

present, particularly in terms of image resolution and arterial conspicuity. Image

subtraction is one of the techniques to improve arterial conspicuity in first-pass

MR angiography. Three subtraction methods can be chosen in practice: cornplex

subtraction, magnitude subtraction and MIP subtraction. This thesis compares

and analyses the effectiveness of these three algorithms under different

situations through cornputer simulations, phantom studies, and clinical studies.

The principles for choosing the appropriate algorithm are provided. Future

techniques to improve the image quality in MR angiography are also discussed.

Acknowledgements

I would like to thank my supervisor, Dr. Graham Wright. I am gratefuf to his

endless patience and support to rny study and Iife.

I would Iike to thank the mernbers of my supervisory cornmittee, Dr. Michael

Wood and Dr. John Rowlands, for their constructive advice and criticism.

I also want to thank al1 the colleagues in our group. Particularly, Christie Webster

for her collaborative works and discussions; Jeff Stainsby for his knowledgeable

answers to al1 my silly questions; and Matthieu Laliberte for his kind help in

improving my English language.

Most of all, I am grateful to rny parents for their support to my study during a

difficult time of the farnily. Hope my mother recovers soon from the surgery and

needs no more contrast-enhanced MRI.

iii

Table of Contents

Introduction ........................................................................................................ vi

... ................................................................................................... Abbreviations viii

Chapter 1 Background: MR Angiography of Peripheral Vascular Disease .. 1

.................................................... 1.1 Clinicat Background: Atheroscierosis in Lower Extremities 1 ......................................................................................................................... 1.1 -1 Pathology 1

1 .1.2 Symptorns ...........................................*....................................................*....................... 2 .................................................................................. 1.1.3 Diagnosis and Treatrnent Planning 3

1.1.4 Treatrnent ......................................................................................................................... 4 1.1.5 X-ray Angiography and MR Angiogrâphy ........................................................................ 5

1.2 Technical Background: MR Angiography .... .. ......................................................................... 7 1.2.1 Conventional MR Angiography .................................................................. ................... 7

1.2.1 -1 Time of Flight Angiography .................... .. .............................................................. 7 1.2.1 -2 Phase Contrast Angiography .................................................................................... 9

............................................................................ 1 .2.2 Contrast-Enhanced MR Angiography 11 ............................. 1.2.2.1 Gharacteristics of Gd-DTPA ... .............................................. 11

........................................................................... 1.2.2.2 Contrast Dose and Injection Rate 12 ...................................................................... 1 .2.2.3 3D Spoiled Gradient Echo Sequence 13

1.2.2.4 Centric Order Acquisition and Contrast Bolus Timing ............................................. 15 ..................................................... 1.2.2.5 Stepping-Table and Multiple-Injection Protocols 16

1.2.2.6 Image Subtraction and Partial Volume Effects ........................................................ 18 ................................................................................. 1 .2.2.7 Maximum Intensity Projection 20

.............................................................................................................. 1.2.2.8 Summary ... 22

Chapter 2 Analysis of Subtraction Methods in 3D Contrast-Enhanced ........................................... MR Digital Subtraction Angiography 23

2.2 Theory ................................................................................................................................... 24 2.2.1 Subtraction ..................................................................................................................... 24 2.2.2 Noise Behavior in MIPs .................................................................................................. 26

..................................................... 2.2.3 Background Tissue Signal Behavior in Subtractions 28

2.3 Materials and Methods ......................................................................................................... 31 2.3.1 Cornputer Simulation ...................................................................................................... 31

iv

2.3.2 f hantom Study ............................................................................................................... 31 2.3.3 Clinical Study ................................................................................................................. 33

2.4 Results .................................................................................................................................. 35 2.4.1 Computer Simulation ......................... .. ....................................................................... 35 2.4.2 Phantom Study ............................................................................................................... 37 2.4.3 Clinical Study .......................... .,. ................................................................................ 40

2.5 Discussion ............................................................................................................................. 44 2.5.1 Arterial SNR and Conspicuity Measurement ................................................................. 44 2.5.2 Partial Volume Effects .................................................................................................... 45 2.5.3 Subtraction in the Stepping-Table Protocol ................................................................... 46 2.5.4 Other VoIurne Rendering Algorithms ................... ,,,, ...................................................... 48 2.5.5 Artifacts by Subtraction .................................................................................................. 49 2.5.6 Do We Need Subtraction in 3D Scans? ............. .. ....................................................... 51

2.6 Conclusion ............................................................................................................................ 52

Chapter 3 Future Directions ........................................................................... 53

3.1 Blood Pool Contrast Agents ................... .. ....................................................................... 54 3.1 -1 NC100150 Injection ........................................................................................................ 54 3.1.2 MS-325 ................... .-. ............................................................................................... 55

3.2 Artery-Vein Segmentation ..................................................................................................... 57 3.2.1 Connectivity-based Algorithms ................... .......... ............................................ 57

....................................................................... 3.2.2 Tem poraI Correlation-based Techniques 58 3.2.3 Phase Contrast-based Techniques ......................................................................... 60

.................................................................................. 3.2.4 Oxygen Level-based Techniques 60

3.3 Summary ............................................................................................................................... 61

Introduction

Contrast-enhanced MR angiography has been applied to virtually al1 vascular

territories in clinical practice. Particularly in the detection of arterial disease in

lower extremities, MR angiography is promising as a replacement of catheter-

based conventional x-ray angiography. To improve the artenal conspicuity in the

MR images, subtraction techniques are comrnonly used to reduce the

background tissue signals as well as the possible venous enhancement. The

effectiveness of subtraction has been discussed in the Iiterature. However,

conclusions drawn in those studies were specific to the different imaging

protocols and subtraction algorithms applied. Generally, cornplex subtraction is

thought to be useful in 2D thick slab projection protocols, and magnitude

subtraction is effective for improving the visibility of small vessels in 3D single-

injectionktepping-table protocols. However, to our knowledge, no work has been

done to evaluate the subtraction techniques in multiple-injection protocols, in

which subtraction is thought to be critical to eliminate the venous enhancement

by the previous injections. In this study, we will evaluate three subtraction

algorithms under a multiple-injection protocol, and discuss the effectiveness of

subtraction under various situations in multiple-injection and single-injection

protocols.

Chapter 1 is a brief introduction to the clinicat background of lower extremity

arterial diseases and the state of the art in MR angiography techniques.

Particularly, previous research on subtraction techniques is discussed.

Chapter 2 is the core of the thesis. The effectiveness of three subtraction

algorithms is derived in theory and verified by cornputer simulation, phantom

studies and patient studies. The considerations for choosing the most appropriate

subtraction algorithm are discussed.

Chapter 3 is a description of future directions for irnproving the image quality of

MR angiography for lower extremities. Of primary interest are two related topics:

steady state blood pool contrast agents and artery-vein segmentation techniques.

vii

Abbreviations

CNR

FOV

Gd-DTPA

MIP

PC

r 1

SNR

T l

T2

~ 2 *

TE

TOF

TR

USPlO

Contrast-to-Noise Ratio

Field of View

Gadolium-Diethylenetriamine-Pentaacetate

Maximum I ntensity Projection

Phase Contrast

Longitudinal relaxivity, per unit concentration of solute, of an agent that alters Tl relaxation rates. r l is expressed in units of (rn M/ 1)-' sec-'

Transverse relaxivity, per unit concentration of solute, of an agent that alters T2 relaxation rates. r2 is expressed in units of (rn MA)-' sec-'

Signal-to-Noise Ratio

Spin-lattice or longitudinal relaxation tirne

Spin-spin or transverse relaxation time

The reciprocal of ~ 2 * is equal to the surn of the reciprocals of the T2 relaxation time and the transverse dephasing caused by fixed magnetic field inhomogeneities.

Echo Time

Time of Flight

Repetition Time

UltrasrnaIl Superparamagnetic lron Oxide

viii

Chapter 1

Background: MR Angiography of Peripheral Vascular Disease

1.1 Clinical Background: At herosclerosis in Lower Extremities

Lower extremity arterial and venous diseases account for significant costs to

society. Arterial disease, commonly known as Critical Limb Ischemia, can

culminate in the loss of a iimb, and patient mortality is frequent, usually the result

of associated cardiac or cerebrovascular pathology' . In the US, about 1 00,000

associaied surgeries are performed each year to relieve symptoms of critical Iirnb

ischemia2. The long terni success rate is over 70%.

1.1.1 Pathology

Atherosclerosis results in the stenosis or occlusion of the arterial lumen by the

plaque formed in the endothelial lining of arterial walls. The exact pathology of

atherosclerosis is still unknown, but evidence now indicates that this condition

begins when the smooth muscle fibres near the tunica interna divide

repeatedly3". Monocytes then invade the endothelial lining and circulating lipids

Chapter 1. MF? Angiography of Peripheral Vascular Disease

accumulate at the endothelium. The result is a lipid plaque projecting into the

inner lumen of the artery (Figure 1.1). When the lumen is occluded, no blood can

flow and oxygen supplies to the associated tissues are Iimited. Most patients

suffenng from symptorns of peripheral atherosclerotic disease are older than 45.

The ratio between male and female patients is about 4:15. Smoking and diabetes

are frequently associated with atherosclero~is~~~. This evidence indicates that

multiple factors may contribute to the formation of atherosclerosis.

Figure 1.1 Cross-sectional illustration of atherosclerosis. The lipid plaque projecting into the inner lumen of the artery lirnits blood flow. (From Martini F. Fundarnentals of Anatomy and Physioiogy. Prentice Hafl, New Jersey, 1992)

1.1.2 Symptoms

In early stages, patients with atherosclerosis in the lower extremities may feel

atypical sensation after a prolonged stay in a fixed position. Intermittent

claudication (pain while walking and exercising) may then appear. As the disease

Chapter 1. MR Angiography of Peripheral Vascular Disease

progresses, patients may feel pain even at rest. For severe disease, ulceration

and gangrene are quite common (Figure 1.2)?

Figure 1.2 Ulceration and gangrene in patients with severe disease due to athrosclerosis in lower extremities. (From Kappert A. Diagnosis of Peripheral Vascular Disease- Bern H. Huber, 1 971)

1.1 .3 Diagnosis and Treatment Planning

The diagnosis of peripheral atherosclerosis is usually a two-step process. First,

patients undergo noninvasive tests, which include a skin appearance check, pain

sensation test, ankle-brachial pressure measurement and probably a Doppler

ultrasound measurement to determine the blood flow. These studies measure the

severity of the disease and help localize the pathology to general anatomic

regionç. More than 70% of atherosclerotic lesions appear in the femoral-iliac

region6. Generally, for half of the patients examined, the noninvasive tests

indicate that the syrnptoms are in fact caused by hemodynamically significant

vascular disease, and they go on to the next level. The second level procedure

typically involves obtaining an x-ray angiogram to more precisely define the

location and extent of diseased vesse1 segments as well as to generate a "road-

map" for the treatment. This generally requires images with a large field-of-view

Chapter 1 . MR Angiography of Peripheral Vascular Disease

(FOV) covering the lower extrernities and high spatial resolution to image the

distal vessels. If significant stenosis or occlusion is found in the angiograrn,

surgical or interventional treatment will be suggested.

1.1.4 Treatment

Besides amputation as the last choice, currently there are two treatment options

for atherosclerosis in the lower extrernities. One is angioplasty, which is the

insertion of a balloon by a catheter and inflation of the balloon to expand the

stenosed artery (Figure 1.3a). Generally a metallic stent will be placed at the

diseased position to support the artety (Figure 1.3b)=. For doing angioplasty, the

length of the diseased region should be measured precisely in the angiogram. An

inappropriate length of stent may result in re-stenosis.

Figure 1.3 a. Angioplasty with a balloon. The stenosed artery is expanded by the inflated balloon. b. A balloon with a metallic stent. The stent will be placed at the diseased position to support the artery after the angioplasty surgery. (From Kim D and Orron DE. Peripheral Vascular Imaging and Intervention. Mosby-Year Book, St. Louis, 1992)

The other option is graft bypass (Figure 1.4). The proximai and distal arteries for

attachment of the bypass should be chosen according to the angiograrn. It is

especially challenging to identify appropriate distal vessels because of the higher

spatial resolution required.


The ratio of angioplasty to graft bypass treatments varies among different centers

(from 1 :4 to 7:3)'. Currently interventional radiologists tend to choose angioplasty

as their first option because of its relatively minimally-invasive nature. However,

under conditions Iike multiple stenoses and calcified atherosclerosis, graft bypass

still works better to date6.

1.1.5 X-ray Angiography and MR Angiography

X-ray digital subtraction angiography is the gold standard for creating vascular

maps. By using contrast agent and digital subtraction techniques, it generates

angiograms with high spatial resolution and high arterial conspicuity (Figure

1.5a). Digital subtraction techniques are applied to improve the arterial

conspicuity of the angiograrns. By subtracting the angiograms before and after

the contrast dye, signals frorn the bones and other tissues are suppressed, and

srnall arteries are better visualized. However, X-ray angiography requires

intraarterial catheterization involving an arterial puncture at the groin and

insertion of a catheter to the illac artery for the injection of the contrast dye. Hours

of hospitalization are needed after the imaging procedure in case of any

complications. Furthemiore, the iodinated contrast dye is nephrotoxic and may

induce acute renal dysfunction in patients with renal disease7. MR angiography is

a newly developed non-invasive modality which is promising as a replacement to

invasive x-ray angiography in the future (Figure 1 .5b)'-12.


Figure 1.4 A bypass graft is a b indicated by the arrow in the X- Figure 1.5 a. X-ray digital subtraction ray angiographie image. (From angiography b. Contrast-enhanced MR Kappert A. Diagnosis of angiograp hy. These techniques provide Peripheral Vascular Disease. comparable visuaIization of the vascuIar Bern H. Huber, 1971) tree of lower extremities. (From Prince

MR et al. Peripheral Vascular MR Angiography Presentation. ISMRM 2000)


1.2 Technical Background: MR Angiography

1 -2.1 Conventional MR Angiography

Conventional MR angiography techniques are attractive due to their non-invasive

nature. Two approaches have been used widely in clinical studies: time of flight

angiography and phase contrast angiography.

1.2.1 -1 Time of Flight Angiography

Time of Flight (TOF) angiography uses a f low-corn pensated gradient echo

sequence. Signals from stationary tissues are saturated by rapidly repeated

excitation pulses (short repetition time (TR)), while signals from blood flowing into

the imaging volume are strong since the magnetization of the fresh blood is not

pre-saturated13. Echo t h e (TE) should be as short as possible in order to

capture the signal before blood flow causes phase dispersion and associated

signal loss. Generally, first-order flow-compensated gradients are applied to

refocus the magnetization of constant velocity blood at the echo time (arrow in

Figure 1.6). Additional gradient lobes can be designed to focus higher order

motion components like acceleration and jerk, but the lengthened TE increases

the effects of even higher order motion components. Flow-compensated

gradients should be chosen based on the flow characteristics of the specific

imaging region.


Figure 1.6 First order fiow-compensated gradient. Signal phase ~ ( t ) = y [x(t)C(t)dt, where x(t)

is the position along the gradient axis over tirne. x(t) = x,, + vt + l /2atz + l /6 jt3 +... ~ ( t ) is magnetic field gradient over time. The magnetization of constant velocity blood is refocused at the

acho tirna (indicated by the arrow); that is, #(TE) = 0, :- p ( t ) d t = 0 , F ~ ( t ) d t = 0

TR should be short to keep stationary tissues saturated but long enough to allow

blood to flow into the imaging plane. At a typical blood velocity IOcrn/sec, for a

3mm slice, the TR needs to be about 30ms in order to have optimal inflow of

fresh, unsaturated blood, assurning the vesse1 is perpendicular to the slice.

A powerful tool for TOF imaging is the pre-saturation pulse which can be placed

above or below the imaging plane to selectively saturate blood from either the

atteries or veins flowing into the imaging plane. In lower extremity imaging,

because the arterial blood generally flows down the legs while venous blood

flows toward the abdomen, a pre-saturation pulse piaced inferior to the imaging

plane can saturate the venous signald3.

TOF imaging can be performed either in 2D slices or 30 volumes. 3D TOF has

higher resolution and eliminates the potential problem of misregistration between

consecutive imaging slices in 2D. However, the thickness of the 3D slab is limited

for sufficient blood inflow, and the contrast of arterial signals and stationary tissue


signals is generally less than for 2D TOF. Therefore, 30 TOF generally is not

used in imaging lower extremities.

One major disadvantage of TOF angiography is that in-plane saturation causes

artifacts associated with slow flow and tortuous arteries or if the long axis of the

vessel coincides with the scan plane. These artifacts look like stenoses of the

arteries. A second drawback is turbulence-induced signal loss in a region of

stenosis which may cause over-estimation of the ~tenosis'~. At these regions,

higher order motion components are dominant and the first order flow-

compensation gradients are less effective. Furthemore, TOF imaging of the

lower extremities takes tens of minutes. Patients' comfort and gross motion are

also concems.

1.2.1 -2 Phase Contrast Angiography

Different from TOF angiography, which relies on the inflow blood for the vessel-

tissue contrast, Phase Contrast (PC) angiography uses signal phase shifts

associated with blood flow in the presence of flow-encoding gradients. In contrast

to TOF sequences, which use flow-compensated gradients to refocus the

dephasing of constant velocity blood, PC sequences emphasize this phase shift

by applying bipolar gradients (Figure 1.7). A second image is acquired by

applying the pair of bipolar gradients with inverted signs. The signals from

stationary tissues are in-phase in the two images while signals from constantly

flowing blood are shifted to opposite phases. Using the phase difference image,


signal phases of blood are proportional to velocity, and stationary background

tissues are s~ppressed'~.

Figure 1.7 Bipolar gradient Gd and its inverse gradient Ge. In Phase Contrast angiography, iwo images are acquired with these two gradients respectively. The phase difference image will highlight the flowing blood and suppress the stationary tissues.

Phase Contrast angiography is only sensitive to a specified range of velocities.

The phase shift of the peak velocity has to be optimized to be less than 180'.

However, in occlusive disease, turbulence causes a broad spectrurn of rapidly

changing velocities. The phase shift from velocities higher than the limit will

exceed 180' and the phase difference is no longer in proportion to the velocities.

On the other hand, if the peak velocity is set too high, images will not be sensitive

to slow flow. Furthemore, the acceleration and higher order motion cornponents

associated with turbulence cause intravoxel phase dispersion and signal 10s~ '~.

Although TOF and PC MR angiography do not need the injection of a contrast

agent, the inherent flow-dependent nature Iirnits their use in clinical patient

studies. A flow-independent technique is preferable for robustness in clinical

p ractice.

Chapter 1. MR Angiography of Peripheral Vascular Disease 11

1.2.2 Contrast-Enhanced MR Angiography

Unlike X-ray angiography, which requires arterial catheterization to inject contrast

agent, MR contrast-enhanced angiography only needs intravenous injection,

which can provide sufficient vesse1 contrast in images. This is considered much

less invasive and therefore is preferred in clinical practice.

1.2.2.1 Characteristics of Gd-DTPA

Gadolinium-based contrast agents are currently the most cornmonly used

contrast agents in MR angiography. ~ d * is a paramagnetic rnetal ion that

decreases both Tl and T2 relaxation times of water in its immediate vicinity.

Because ~ d ~ + itself is biologically toxic, it is chelated with ligands such as DTPA

(diethylenetriarnine pentaacetate) to f o m a small-molecular contrast agent15.

These extracellular agents diffuse from the intravascular compartment into the

interstitial space in a matter of minutes (except in the brain because of the brain-

blood barrier), so selective imaging of the vascular structures must be performed

rapidly.

At 1.5T, the relationship between Tl and Gd concentration can be approximated

as8


where r l is the longitudinal relaxivity of the gadolinium chelate. At 1.5 T, r l is

approximately 4.5~-~(mm01/1)-'. T l of blood without contrast agent is 4230ns8.

When Gd concentration is 2mrnol/l, blood T l can be shortened to about 100ms,

which is significantly shorter than that of surrounding muscle (T1=800ms) and fat

(Tl =270ms). By applying a Tl -weighted pulse sequence, the tissues with shorter

T l will have a stronger signal in the image. Thus, blood with contrast agent can

be differentiated from the surro unding tissues.

1.2.2.2 Contrast Dose and Injection Rate

From an image quality point o f view, generally more contrast is better. However,

it is important to consider the issues of safety and cost. In clinical studies, doses

between 0.1 to 0.3mmoV(kg patient weight) are generally used8 (e.g. for a patient

of 65kg weight, 15-40ml of contrast agent at O.5mmoVml will be applied). Unlike

X-ray angiography in which the contrast dye is injected intraarterially and the X-

ray image is then captured repeatedly, in contrast-enhanced MR angiography,

the contrast is injected intravenously from the forearm and it takes tens cf

seconds for the contrast bolus to arrive at the imaging FOV. The contrast bolus

tends to lengthen as it travels from the cubital vein through the heart and lungs to

the artenes being imaged. Th is contrast dispersion has an important effect on

image signal-to-noise ratio (SNR). Furthenore, 3D MR imaging may take from

several seconds to more than a minute, depending on the image resolution.

Therefore, contrast injection rate should be chosen based on the location of the

volume of interest and the timing parameters of the pulse sequence. Generally,


injection rates of 0.5-2mVs are used in petipheral vascular imaging with a saline

flush immediately following at the same rate8. A fast injection rate will increase

the concentration of contrast agent within the vessels, but venous enhancement

is more likely to occur, especially during imaging of the distal region. A slow

injection rate can avoid the venous enhancement problem due to the leakage of

the contrast agent into the extracellular space during the first pass through the

capillary bed, but lower SNR can be expected due to lower concentration.

1.2.2.3 3D Spoiled Gradient Echo Sequence

A 3D spoiled gradient echo sequence is most commonly used in contrast-

enhanced MR angiography because of its Tl weighting, high speed, and short

echo time. Without the requirements of a slice-selective gradient and RF pulse in

2D imaging, 3D imaging rninimizes stress on the gradients and allows the use of

shorter RF pulses. These factors allow the use of shorter repetition times and

thinner sections. Section thickness can be reduced to under 1 mm provided there

is sufficient SNR. The shortened repetition and echo times make it possible to

collect large 3D volumes of data in tirnes on the order of a minute. This high

speed is critical in first-pass measurement of the contrast enhancement.

Furthemore, 3D data sets can be oriented and reformatted in any desired plane

in post processing, which provide more useful information than 2D protocols.


Figure 1.8 30 spoiied gradient echo sequence, consisting of excitation pulse (RF), spatial encoding gradients (Gx,Gy,GJ, time of acquisition or echo time (TE), and sequence repetition time (TR), The residual transverse magnetization after data collection is destroyed by the RF pulse. The phase of the RF is varied from one excitation to another to avoid build-up of transverse coherence.

TE should be short enough to eliminate dephasing artifacts and to minimize ~ 2 "

signal decay (Figure 1.8). This requires an echo time less than about 3ms8.

Shortening TR translates directly into shorter data acquisition times. The resulting

decrease in the SNR can be compensated by tightening the contrast bolus with a

faster injection rate and by adjusting the flip angle of excitation to the expected

blood Tl. In general, the TR should be made as short as possible without

excessively increasing the receiver bandwidth. A TR less than 10ms is preferred

Gradient echo imaging requires selection of the flip angle of excitation. The

optimized Ernst angle depends on the TR and contrast concentration


(8 = cos-' [e-"'"]). In general, for lower contrast doses and very short TR, a lower

flip angle rnay be optimal. For higher doses and longer TR, a higher flip angle

may be more appropriate8. Flip angles generally are chosen in the range of 20-

60°.

Spoiling the residual transverse magnetization after each echo is useful because

it establishes a robust steady-state magnetization insensitive to small system

variations. Figure 1.8 is a typical 30 spoiled gradient echo sequence. The phase

of the RF pulse is varied from one excitation to another in order to avoid coherent

16,17 buildup of the transverse magnetization .

1.2.2.4 Centric Order Acquisition and Contrast Bolus Timing

Because the centre of k-space or low spatial frequencies dominate image

contrast, whereas the periphery of k-space or high spatial frequencies contribute

more to fine details such as edges, central k-space acquisition is generally timed

to when the contrast agent first arrives in the arteries of interest to highlight

arterial in fornat i~n '~ '~~. This ensures higher arterial SNR and maximum artery-to-

vein contrast. To acquire the central k-space data when contrast is distributed

unifomly in the arteries, timing of the contrast arrivai to the volume of interest

becomes a critical concern. An early start of the scan will lose arterial signal while

a delayed start may sacrifice artery-to-vein contrast. The blood velocity varies

significantly among patients. Thus specific bolus detection techniques are

preferred over the estimation of arriva1 times based on experience.


A 1-2cc test bolus can be used prior to the actual scan to help estimate the

arrival time2'. This bolus is flushed with a sufficient volume of saline to mimic a

full injection. A series of 2D fast gradient echo images of the appropriate vascular

region are then obtained as rapidly as possible (-1 slice/sec) for about a minute.

The time of peak arterial enhancement is then used as the arrival time estimate.

However, the irnaging bolus may not behave in a manner identical to the test

bolus because of patient variables such as venous return and cardiac output.

A more sophisticated approach involves directly monitoring a vascular structure

for arriva! of the full bolus contrast material and then automatically triggering the

centric acquisition once the signal intensity exceeds a t h resho~d~~ -~ . This

provides the most reliable way of ensuring acquisition of the central k-space data

during the moment of peak arterial phase contrast concentration. The main

problem of this technique is that patient motion rnay cause the tracked volume to

be no longer aligned with the same vascular structure8.

1.2.2.5 Stepping-Table and Multiple-Injection Protocols

Because of the large FOV of the lower extremities, generally the study is divided

into two to three consecutive imaging stations. One option is to image the

consecutive stations following a single injection of contrast agent. Due to the

relatively short time of the arterial phase during the first pass of the contrast

agent (less than 1 minute), quick shifts between the consecutive stations are


preferred. A specially programmed stepping table can quickly and precisely shift

the patient's position in a penod on the order of seconds and two to three stations

can be scanned consecutively in about one minute following one injection of

contrast agen?5-27. However, the scan time for each station is at most 20

seconds, which Iimits the spatial resolution that can be acquired. Furthemore, to

avoid venous enhancement at the distal station, a slow contrast injection rate has

to be applied. This also Iimits the contrast concentration and thus SNR of the

arterial signals. Generally, single-injectionlstepping-table protocols can acquire

fairly satisfying image quality at the renal and illiac-femoral stations, but yield

25-26 inferior quality at the distal vessels in the lower legs .

The other option is to apply a separate volume of contrast agent for each imaging

Without worrying about the contrast left for the subsequent stations, a

longer scan time is available and a faster injection rate can be applied. This

results in a better image spatial resolution and a better SNR than stepping table

protocols. The problem of multiple-injection protocols is the venous enhancement

resulting from the accumulated contrast agents introduced during the earlier

station scans. Usually subtraction or segmentation techniques have to be applied

to eliminate the venous signals and increase arterial conspicuity. However, an

important characteristic of multiple-injection protocols is that we can choose the

order of the stations to be scanned. If our concem is primarily the distal small

arteries, we can scan the lower legs first. Because it is the first station to be

scanned, there is no venous contamination by previous contrast agent.

Therefore, an angiogram with higher resolution and SNR as well as better artery-


vein conspicuity can be achieved for the distal arteries compared with that in

single-injectionlstepping-table protocols. This is the most attractive feature of

mu ltiple-injection protocols.

1.2.2.6 lmage Subtraction and Partial Volume Effects

30-36 lmage subtraction is cornmonly used to suppress background tissue signais .

Particulariy, in multiple-injection protocols, subtraction is useful to eliminate

residual venous signals. It has been demonstrated that subtracting complex raw

data is useful to reduce partial volume e f f e ~ t s ~ ~ ; that is, contributions of signal

frorn tissue outside of the vesse!. This can be illustrated by Figurel.9:

lmage Pixel

essel tissu

a Figure 1.9 Illustration of paniai volume effects and its elimination by complex subtraction. In a, the image pixel size is larger than the vessel size. The signal intensity of the image pixel is the sum of the vessel signal and the tissue signal. This is called partial volume effects. In b, the - - image pixel signal before contrast enhancement Mo is the sum of the vessel signal rn, and the - tissue signal rn, . After the contrast enhancement, the tissue signal does not change. The vessel - - signal is enhanced to m, . The image pixel signal after enhancement is then Mc . If we subtract - - the magnitude of Mc and M o , we get near zero signal. However, if we subtract in the complex - - domain, that is, subtract the vectors M c and M o , we then get the real difference of the vessel


As shown in Figure 1.9 a, when the image pixel size is larger than the vessel

size, it includes not only the vessel signal but also the signal from tissue

surrounding the vessel. The signal intensity of the image pixel iç the sum of t h e

two parts. This partial volume effect prevents the precise measurement of t h e

vessel signal intensity.

In Figure 1.9 b, the image pixel signal Mo in the mask data set is the sum of t h e

- vessel signal and the tissue signal rn, . With the enhancernent of the contraist

agent, the vessel signal increases to . It should be stressed that the contraist

agent not only enhances the magnitude of the vessel signal, but also affects iits

phase. The net signal of the pixel is then K. If we subtract the magnitude of t h e

two signals Iz[ and 1x1, we may get a near zero result (as illustrated). However,

if we subtract the cornplex data, that is, the vectors c a n d Mo, we will get t h e

- - - result m, . rn, equals the difference of the two vessel signals and rn, .

Therefore, subtracting the complex numbers can eliminate the partial volume

effects caused by the tissue signal in the image pixel.

Complex subtraction works efficiently in 2D thick slab projection and I a w

resolution 3D imaging protocols in which partial volume effects are significant.

However, in high resolution 3D protocols such as the one we are using, t h e

image pixel volume is about 0 .8~0 .8~1 mm in the lower legs, and 1.6~1.6~1 mm in

the upper legs, smaller than the vessels in the corresponding regions. In t h e

image pixels, there are therefore either only vessel signals, or only tissue signals.


Partial volume effects are effectively avoided. As we see in Figure 1.9 b, if there

- are only vessel signals m, and z, their magnitude difference also can reflect

the contrast enhancement. Therefore, subtracting complex numbers is not critical

in Our protocol. On the contrary, we will demonstrate in Chapter 2 that, in terms of

background suppression, cornpfex subtraction perforrns poorly compared to other

algorithms.

1.2.2.7 Maximum lntensity Projection

To present the 3D data sets on cornputers and films, volume rendering

techniques are applied to transfomi the 3D data sets into 2D displays. In

contrast-enhanced MR angiography, maximum intensity projection (MIP) is the

most widely used aigorithm13. This algorithm picks the maximum intensity pixel

along each projection ray as the corresponding pixel in the 2D image. Because

the signal intensities of contrast-enhanced blood are greater than its surrounding

tissues, the higher blood signals in the projection plane should be retained. This

algorithm is fast and operator independent, and provides higher SNR and

contrast-to-noise ratio (CNR) compared with a single slice image37-38. MIP can be

perfortned in any orientation. Thus the data set can be interpreted from multiple

angles.

Despite its usefulness, the MIP algorithm is subject to artifacts. First, there is no

depth cue information. Thus two adjacent vessel pixels in the resulting image do

not necessarily belong to the same branch of the vascu~ature'~. This overlapping


effect may cause confusion in depiction of vascular trees. MlPs from multiple

angles may help to solve this problem.

However, the biggest concem with MIP is its effect on the increase of the

background signal level. Compared to single slice images, the mean of the

background distribution is shifted higher in the MIP images in proportion to the

length of the projection rays37-38. Generally vessel edges have only slightly higher

signal intensities than the average signal intensities of surrounding tissues, which

is caused by partial volume effects at the edges. After the MIP processing, this

vessel edge information and low intensity small vessels may be replaced by the

background tissue signals in the MIP image. This results in narrower vessel

appearance compared to the single slice images. In clinical studies, this may

cause overestimation of a stenosis13. To overcome this problem, a local MIP can

be applied (only projecting at the vessel signal reg ion^)^'^'. This is generally

achieved by a threshold processing in advance to segment the vessel signals

and nuIl the background signals. Some vessel connectivity algorithms have also

been tried for the segmentation4149. However, because of the variability of the

vessel and tissue signals in MR images, there is always overlap between the

vessel and tissue signal intensities. Sorne small vessels are usually lost after

segmentation. This limits the application of local MIP. Sum projection does not

work well in MR angiography because of the increase of background variance

and the averaging of vessel signals with background tissue signalsI3. To date,

the whole volume MIP is still the most widely used algorithm to present 3D MR

angiography data.


1.2.2.8 Summary

Contrast-enhanced MR angiography has been demonstrated to be a reliable

app roach for lower extremity irnaging in dinical practice&12. Various tech niques

need to be investigated further to optirnize its performance, especially in ternis of

spatial resolution and arterial conspicuity. This thesis is an investigation of one of

the critical approaches for improving the image quality in MR angiography: image

subtraction techniques.

Chapter 2

Analysis of Subtraction Methods in 3D Contrast-Enhanced MR Digital Subtraction Angiography

2.1 Purpose

Subtraction is a commonly used technique to increase arterial conspicuity in

contrast-enhanced MR angiography. Although in most papers the subtraction

methods were not stated, generally magnitude images reconstructed by the

scanner were u ~ e d ~ @ ~ ' . In the last few years, the importance of complex

subtraction has been identified36 (see section 1.2.2.6). In theory, subtracting

complex numbers can reduce partial volume effects and recover some vesse1

signals. However, we found complex subtraction did not always give us the best

result in 3D MR angiography. On the contrary, subtracting two MIP images,

which was thought to be hot usefu~'~, often yielded the best arterial conspicuity

(Figure 2.1). This suggests that different subtraction algorithms should be applied

under different situations. An anaiysis of the underlying reason for different

results through mathematical modeling and clinical data evaluation is needed. To

Our knowledge, this work has not been done in the past. This study will provide a

Chapter 2. Analysis of Subtraction Methods in 3D CE MR DSA 24

theoretical basis for choosing the most appropriate subtraction algorithm for MR

angiograp hy.

Figure 2.1 Clinical example of different subtraction algorithms in 3D MR angiography. Grayscale table is inverted to better visualize low signai intensity (small) arteries. a. unsubtracted; b. after cornplex subtraction; c. after magnitude subtraction; d. after MIP subtraction. The arterial branch (arrow) is better visualized after MIP subtraction.

2.2 Theory

2.2.1 Subtraction

Two characteristics of the MR data sets contribute to the multiplicity of choices

for subtraction algorithms. First, MR raw data are complex numbers. We can

therefore either subtract the complex numbers of the two corresponding data sets

pixel by pixel then take the magnitude, or we can first take the magnitude of the

two data sets then subtract these magnitude values. In MR imaging, the

difference of order could be significant in the presence of partial volume effects.

The second characteristic is that we use a three-dimensional acquisition to cover

the vasculature. Thus, some kind of volume rendering algorithm is needed to

Chapter 2. Analysis of Subtraction Methods in 3D CE MR DSA

convert the 3D data set into 2D f o m to be displayed on a computer screen or

film. MIP is the rnost popular volume rendering algorithm in 3D MR angiography

because it is operator independent, no thresholding is involved, and it is relatively

fast (see section 1.2.2.7). Although it has sorne potential problems, it seems no

other algorithm is currently more practical and objective in the overall sense. We

can either first do the subtraction of the two 30 data sets then do the MIP to

show the subtracted image, or first do the MIP of the two 3D data sets then

subtract the two MIP images.

Combining these two characteristics, actually the 'subtraction methods' include

both the process of subtraction and the way to show the 2D result. There are

three algorithms that can be chosen:

Complex subtraction: subtract the cornplex numbers of the two 3D data

sets; then take the magnitude and do MIP to get the resultant image;

Magnitude subtraction: take the magnitude of the two 3D data sets first and

do the subtraction; then do MIP to get the resultant image;

MIP subtraction: take the magnitude of the two 3D data sets and get their

MIP images respectively; then subtract the two MIP images to get the

resultant image.


2.2.2 Noise Behavior in MlPs

The vesse1 signal intensities after each subtraction algorithm remain similar.

Therefore, the visualization of low intensity vessels is mainly detennined by the

effectiveness of background tissue suppression. As indicated in Section 1.2.2.7,

MIP wili increase the background tissue signal level. To understand this effect

quantitatively, we first develop the background noise behavior in MIP.

ldeally both the real and imaginary components of the air noise signal in complex

MR data are Gaussian di~tributed'l-~~, with zero mean and standard deviation a.

We can normalize the standard deviation to 1 which yields the probability density

function:

x is a random variable describing amplitude of real or imaginary component.

After taking the magnitude of the complex values, the noise distribution becomes

f3ayleighs1 :

x , , x,are random variables describing Gaussian noise in real and imaginary components.

The accumulated probability function is then:


Suppose the projection length of a MIP is N pixels. The probability P,(z) that al1

the intensity values along the ray are less than is

1 If we define a MIP background level at z, where P, (z,)=-, which is the median

2

of the distribution, then

so that for

and for

In a Gaussian-like distribution, the rnedian is approximately the same as the

rnean. We can therefore plot the relationship between the increase of the air

noise rnean versus the length of the projection ray in Figure 2.2. In our clinical

protocol, where the projection length is 60 pixels, the background noise mean is

about 2.5 times higher than the noise rnean for a single slice magnitude image.

When doing multi-angle MIPs, the projection length can be 256 or 512 pixels and

the increase of background mean will be even more significant.


Figure 2.2 Mean of the background air noise distribution shifts in proportion to the length of the projection ray.

2.2.3 Signal Behavior in Subtractions

In complex subtraction, we subtract two Gaussian distributed air noise signals

with zero mean and standard deviation normalized to 1. The difference signals

remain Gaussian distributed with standard deviation fi. After taking the

magnitude and doing the MIP for N=60, the mean of the distribution is 2.99x&.

In magnitude subtraction, we take the magnitude of the two Gaussian

distributions before subtraction. They become Rayleigh distributed with mean

1.253 and standard deviation 0.655~'. Af€er subtracting these two Rayleigh

distributions, the mean of the signals is zero, with standard deviation 0.655xA.

The subtracted signals are approximately Gaussian distributed. So, after doing

the MIP, the mean of the distribution is 2.3 times the standard dev ia t i~n~~ , that is,


2.3x0.655xZ. Comparing the means in air regions for the two subtraction

methods:

That is, after cornplex subtraction, the mean of the air noise signals is twice that

of magnitude subtraction.

In non-zero signai regions, signals are Rician distt- ib~ted~~-~~. In the signal

regions where SNRr5, the noise is again approximately aussi an^'. In our

imaging protocol, the noise distributions of the artet-ial and venous regions as well

as the background tissue regions in the complex raw data are approximately

Gaussian. In complex subtraction and magnitude subtraction, the background

tissue signals are subtracted to yield zero mean before performing the MIP, so

the signal behavior in background tissues is similar to that of the background air

noise (Figure 2.3). That is, after doing the MIP, the mean of the background

tissue signals also shifts to about 2.5-3 times its standard deviation. On the

contrary, in MIP subtraction, the rneans of the tissue signal distributions in the

two MIP images are both shifted higher before subtraction, and when subtracted

yield a zero rnean. Therefore, the background tissue level after MIP subtraction is

significantly lower than that of both complex and magnitude subtractions. In the

arterial and venous signal regions, although the signal means also tend to

increase after the MIP, the vessel size along the projection ray is only several

pixels. The shifts of vessel signal distribution are not as significant as that of the


background tissue signal distribution. This results in the lower artery-tissue

conspicuity after complex subtraction and magnitude subtraction.

Background Tissue Signal Distribution in Subtractions " ~ a f

1 / \ -agnai Unsubtracted MIP lntensity

complex subtraction

MIP subtraction

Figure 2.3 Illustration of the background tissue signal distributions in subtractions. In the complex subtraction, two Gaussian distributed 3D data sets are subtracted. Magnitude values are then taken and MIP is done. The mean of the signal distribution is shifted higher afier the MIP processing. In the magnitude subtraction, two Rician distributed 3D data sets are subtracted. After doing the MIP, the mean of the signal distribution is also shifted higher. In MIP subtraction, two MIP images are subtracted. The subtracted image is already two-dimensional. No further MIP processing ensures the mean of the distribution is the smallest of the three subtraction methods.


2.3 Materials and Methods

2.3.1 Computer Simulation

Computer simulations were done to verify the theoretical derivation of

background air noise and tissue signal distributions dunng subtraction and MIP

processing. Two Gaussian-distributed cornplex 3D data sets with zero mean and

standard deviation 1 were subtracted to simulate the behavior of background air

noise. Background tissue signals have higher signal intensities and bigger

variance than air noise signals. According to a rneasurement of the clinical

images, two Gaussian-distsibuted 30 data sets with mean 14 and standard

deviation 2 were subtracted to simulate background tissue signals. Data

distributions in the resulting 2D data sets were caIculated.

2.3.2 Phantom Study

A vesse1 phantom was made to verify the effectiveness of the subtraction

methods. Arterial SNR, artery-vein conspicuity and artery-tissue conspicuity are

easier to measure in a phantom study than in clinical cases due to more unifom

signal distributions in a phantorn. These measurements are the direct reflections

of the effectiveness of the subtraction algorithms. We define the following three

parameters:

Arterial Signal Intensity Arterial S N R =

Noise Standard Deviation in the Arteries


Arterial Signal Intensity - Venous Signal Intensity Artery - Vein Conspicuity =

Arterial Sipal Intensity (2-5)

Arterial Signal Intensity - Mean of Tissue Signal Intensities Artery -Tissue Conspicuity =

Artenal Signai Intensity (2-6)

We believe this definition of arterial conspicuity (contrast sensitiviV0) makes

more sense than the CNR measurements in the previous papers 30-35 as the

evaluation of the effectiveness of subtractions. The detaiis are discussed in

Section 2.4.2.

The artery-vein phantom consists of two plastic tubes of diameter 2.5rnrn in a

cylindrical container filled with distilled water. The two tubes were both filled with

Gd-DTPA at a concentration 0.95mmol/l (T1=200ms) to sirnulate an artery and

vein in the steady state of contrast enhancement. A centric ordered 3D Spoiled

Gradient Echo sequence (section 1.2.2.3 and 1.2.2.4) was applied for data

acquisition on a 1.5T GE Signa scanner using the body coil, FOV=20~2OxGcm,

TF(TTE=9.5/1.5ms, flip angle=20°, 256~256x64 matrix. The Gd in the arterial tube

was then drained and a higher concentration of Gd at 2.lOmmoVI (T1=100ms)

was injected instead to simulate an artery in the first-pass state of contrast

enhancement. Another 3D scan then was performed. The two data sets were

processed for subtraction by al1 three algorithms and signal behaviors were

measured numerically. A single pixel in the arterial tube and a single pixel in the

venous tube were picked as the artenal and venous samples. A 20x20 pixels


square region in the uniform water section was selected as the background

tissue sample. The variance of the background air noise in the unsubtracted MIP

image was used to estimate the noise in the artenal regions (see section 2.4.2 for

details).

2.3.3 Clinical Study

The effectiveness of subtraction rnethods was also evaluated in clinicat studies.

Five patients with suspected peripheral vascular disease were scanned on a 1.5T

MR system (GE Signa Advantage 5.7) using a two station, two injection protocol.

Two separate IV injections of a 20ml gadolinium chelate contrast agent

(Magnevist, Berlex) with a 20ml saline flush were applied for lower legs and

upper legs respectively, at an injection rate of 1 -5mVs. To emphasize image

quality in the distal vessels, we did the lower leg scan first. A 2D TOF Iocalizer

scan was applied to prescribe the FOV of interest of one lower leg. The first bolus

of contrast agent was then injected. After a timed deIay, the prescribed FOV was

scanned in the first-pass arterial phase with the same 3D pulse sequence used in

the phantom studies, but with half k-space acquisition to get more high resolution

information, FOV=40x20x6cm, 5 1 2~256x64 matrix. Another 2D TOF localizer

scan was then applied for the upper legs. During this time, the contrast agent

from the first injection had become unifomly distributed in the legs, so both the

artenal and venous signals as well as background muscle signals were

enhanced. To do the subtraction, the upper legs were scanned by the 3D pulse

sequence prior to the second injection as the mask data set (FOV=40x40x6cm,


256x256~64 matrix). After the mask scan, the second contrast volume was

injected and the same scan was repeated after a timed delay. We assumed the

contrast concentration in veins did not change much between these two scans,

so subtraction would reduce venous signals to near zero.

All of the three subtraction algorithms were perfomed on a Sun Workstation. To

evaluate the effectiveness of background signal suppression, the statistics of the

signal distributions in the background leg tissues were calculated. To avoid the

statistical variability caused by the variance of the tissue signal distributions, the

statistics of the air noise region was also calcuiated as a reference. A 13x13

pixels square region outside of the body was selected as the background air

sample (A in Figure 2.7). A region of the same size in the legs with no obvious

vesse1 signals was selected as the background leg tissue sample (T in Figure

2.7). Signal intensities of the three subtraction images were normalized according

to a selected arterial signal pixel. Means and standard deviations were calculated

and norrnalized to the mean of complex subtraction.


2.4 Results

2.4.1 Cornputer Simulation

The computer simulation was used to verify the theoretical derivation. Figure 2.4

shows the background air noise distributions before and after the three

subtraction algorithms. After complex subtraction, the mean of the distribution is

even larger than that of the unsubtracted data. This is caused by the increase of

signal variance after subtracting Gaussian-distributed cornplex numbers. The

mean after MIP subtraction is significantly lower than that of the other two.

Figure 2.5 shows tissue signal distributions before and after subtraction. After

complex and magnitude subtraction, the background tissue levels remain about

1/3 to If2 of the un-subtracted level (depending on the initial SNR). After MIP

subtraction and taking the absolute value, the background tissue level is much

lower. If the arterial signal intensities afier subtraction fall into the region indicated

by the arrow in Figure 2.5, they will be lost after complex and magnitude

subtraction, while they will survive after MI P subtraction.

Table 2.1 and Table 2.2 list the relative mean and standard deviation of the

signal distributions in Figure 2.4 and 2.5, respectively. As expected from Section

2.2.3, the air noise mean for magnitude subtraction is half that for complex

subtraction. The mean values after MIP subtraction are significantly smaller than

those after complex and magnitude subtractions.


Figure 2.4 Background air noise distributions in computer simulation. The mean of the distribution after complex subtraction is even larger than that before subtraction. The mean after MIP subtraction is significantly smaller than that of the others.

Figure 2.5 Background tissue distributions in computer simulation. The mean of the distribution after complex subtraction is slightly larger than that of magnitude subtraction, while both of them are much larger than that of MIP subtraction. If vesse1 signals after subtraction fall into the region


indicated by the arrow, they wiII be lost after complex and magnitude subtraction, white they wiH survive after MIP subtraction.

Table 2.2 Statistics of Background Tissue in the Computer Simulation 1 Com~lex Sub 1 Maa Sub MIP Sub 1

Table 2.1 Statistics of Background Air Noise in the Computer Simulation

1 Tissue Mean 1 1 1 0.76 1 0.1 2 1

Air Noise Mean Air Noise Std Dev

! Tissue Std Dev f 0.13 1 0.15 1 0.09 1

2.4.2 Phantom Study

Compiex Sub 1

0.1 3

Images from the phantom study are shown in Figure 2.6. It is clear that cornplex

subtraction has the worst artery-background conspicuity, and MI P subtraction

has the best. The statistics of background air noise and tissue signals in Table

2.3 and Table 2.4 are in general agreement with those of the computer simulation

in Table 2.1 and Table 2.2. In Table 2.5, the artery-vein conspicuity is improved

after al1 three subtraction algorithms. However, the artery-tissue conspicuity after

complex and magnitude subtraction is even worse than that of the unsubtracted

image, but is significantly improved after MIP subtraction. As expected, the

arterial SNR is reduced after al1 subtractions since the arterial signal is subtracted

out and the noise variance is increased. In conclusion, MIP subtraction generates

the best arterial conspicuity at the expense of losing arterial SNR.

Mag Sub 0.50 0.10

MIP Sub 0.1 0 0.08


After MIP subtraction, the venous tube is still slightly visible. This can be

explained as follows: the variance of the venous signals is bigger than that of the

surrounding water signals in the unsubtracted MIP image. After MlP subtraction,

although the rnean difference was subtracted out, the variance difference still

existed. However, because the mean of the rernaining venous signals is rnuch

lower than that of arterial signals, it will not cause significant problems to artery-

vein differentiation. A threshold can be applied to remove the remaining veins

from the image. In both complex subtraction and magnitude subtraction, most of

these remaining venous signals were replaced by tissue signals in the final MIP

images due to the raised background tissue level. Because the artery-tissue

conspicuity of complex and magnitude subtraction is lower than the artery-vein

conspicuity of MIP subtraction, it is harder to threshold the raised background

tissue in the cornplex and magnitude subtraction.


Figure 2.6 Phantom study. a. mask MIP image; b. contrast-enhanced MIP image before subtraction; c. MIP image after complex subtraction; d. MIP image after magnitude subtraction; e. after MIP subtraction.

Table 2.3 Statistics of Background Air Noise in the Phantom Study

[ Tissue Std Dev 1 0.1 4k0.01 0.1 6kO.02 0.1 2rf-0.01 1 Means and standard deviations of two phantom studies were measured separately and were

Table 2.4 Statistics of Background Tissue in the Phantom Study

normalhed to the mean after the complex subtraction. The f 1 standard deviations of the two measurements are given in Table 2.3 and Table 2.4.


Mag Sub 0.56H.02 0.1 1 H.01

Complex Sub 1

0.14+0.01

Tissue Mean

MIP Sub 0.1 3kO.01 0.1 OkO.01

Mag Sub 0.80+0.01

Complex Sub 1

Table 2.5 Conspicuity and SNR Statistics in the Phantom Study

MIP Sub 0.1 7kO.01

Before Sub 0.36 0.67 18.80

MIP Sub 0.84 0.91 6.1 8

Mag Sub 0.55 0.60 6.1 8

Artery-Vein Conspicuity Artery-Tissue Conspicuity

Arterial SNR

Complex Sub 0.51 0.49 6.1 8


2.4.3 Clinical Study

Figure 2.7 Clinical case 1. Upper leg images before and after subtractions. a. mask MIP image; 6- contrast-enhanced MIP image before subtraction; c, MIP image after complex subtraction; d. MIP image after magnitude subtraction; e. after MIP subtraction. Square region A in a: air sample; square region T in a: tissue sample; square region Z in b: magnify to Figure 2.8.

A set of clinical scan images is shown in Figure 2.7. They are al1 displayed using

full gray scale; that is, no thresholding is used. In mask image a, the contrast

injected earlier for the lower leg scan has already been evenly distributed in the

arteries and veins. In image b, the arteries were enhanced by the first pass of the

freshly injected contrast agent, while the veins were still enhanced from the

earlier contrast. Without subtraction, although the arterial signal intensities are

higher than venous signals, their difference is not sufficient to be discemed by

hurnan vision, especially the superficial femoral arteries and veins. The MIP


image after complex subtraction is shown in image c. Although the interference of

veins was significantly reduced, some arterial branches were immersed in the

signals from the background tissues and cannot be identified. Image d is the MIP

image after magnitude subtraction. The contrast between arteries and

surrounding tissues is better than complex subtraction. However, image e, which

is after MIP subtraction, gives the best background suppression and arterial

depiction. To better illustrate the difference, a magnified square region Z of

Figure 2.7 is shown in Figure 2.8.

Figure 2.8 Magnified region from Figure 2.7. Grayscale table is reversed to better present small vessets. a. unsubtracted; b. after cornplex subtraction; c. after magnitude subtraction; d. after MIP subtraction.

in the unsubtracted image a, the arteries are blurred by the overlapping venous

signals and surrounding tissue signals. After complex subtraction, the high signal

Chapter 2. Analysis of Subtractiion Methods in 3D CE MR DSA

intensity proximal part of the vesse1 is better presented in image b, but the two

low signal intensity branches are often immersed in the background and are

poorly depicted. On the contraty, these two branches are well displayed in image

d, which is after MIP subtraction. Maignitude subtraction yields a result between b

and d.

The statistics of the five clinicai cases are shown in Table 2.6 and Table 2.7.

They are in general agreement with the statistics in the computer simulation. We

assume the slightly larger nurnbers in clinical air noise statistics are caused by

the hat k-space acquisition used in clinical study, since other imaging

parameters are the same as those I n the phantom study, where statistics are in

agreement with the computer simulation. The larger numbers for MIP subtraction

in background tissue statistics may also be due to some background structures

falling into the rneasurements.

Table 2.6 Background Air moise Statistics in the Clinical Study 1 Com~lex Sub 1 Maa Sub 1 MIP Sub


Table 2.7 Background Tissue Statistics in the Clinical Study

0.1 5M.01 0.1 2k0.01

I Y

Tissue Mean Tissue Std Dev

1 0.1 3k0.01

Means and standard deviations of five patient studies were measured separately and were norrnalized to the rnean after the complexz subtraction. The tl standard deviations of the five measurements are given in Table 2.6 and Table 2.7.

Complex Sub 1

0.1 6k0.02

Mag Sub 0.76+0.02 0.1 6M.02

0.68k0.03 0.1 4H.01

MIP Sub 0.23k0.09 0.1 8k0.08


However, we also observed an exceptional case in Figure 2.9 in which MIP

subtraction perfomed worse than complex subtraction and magnitude

subtraction. After complex subtraction and magnitude subtraction, there is a

longitudinal srnall artery (arrows), which cannot be seen after MIP subtraction.

This is because the signal intensities of this artery are lower than the surrounding

tissues and thus were replaced in the MIP image a before subtraction. On the

contrary, in complex subtraction and magnitude subtraction, the surrounding

tissue signals were subtracted out pixel by pixel before doing the MIP, and those

low arterial signals finally ernerged from the background with the MIPs. However,

this situation is very rare in our clinical studies. This case is probabiy caused by

an underestimation of scan timing after the second injection of contrast agent,

which can be rectified in future studies with new autornatic triggering methods.

a b c d Figure 2.9 Another magnified region from clinical case 2. a. unsubtracted; b. after complex subtraction; c. after magnitude subtraction; d. after MIP subtraction.


2.5 Discussion

Besides the effectiveness of background tissue suppression, there are other

considerations when choosing the appropriate subtraction algorithrn, including

the arterial SNR of the imaging protocol, the partial volume effects, and the

potential artifacts after subtractions.

2.5.1 Arterial SNR and Conspicuity Measurement

Due to the small size of the arteries, it is hard to select a uniform arterial region to

directly measure the arterial noise distribution. To calculate the arterial SNR in

the MIP image before subtraction, we select an arterial pixel manually and

assume it is the only arterial signal along its projection ray. This implies that the

signal distribution of the arterial pixel is not altered by the MIP processing. We

then measure the rnean signal intensity in an air region and use the relationship

in Figure 2.2 to calculate the original noise standard deviation in the complex raw

data. This noise standard deviation is used as an approximation to the arterial

noise in the MIP images before subtraction. After each of the three subtractions,

the arterial signal is reduced a similar amount, and the noise standard deviation

is & times bigger than in the original image. Therefore, the arterial SNRs after

each of the three subtractions are reduced roughly equally.

To evaluate arterial conspicuity, we believe that the measurement of contrast

sensitivity (equation 2-5, 2-6) is more meaningful than the measurement of CNR.

The contrast sensitivity measures the relative difference between two signal


levels and assumes the noise levels of the two signals are smaller than the

difference (which is true in Our cases). This measurement actually reflects the

relative gray scale level difference of the two signals on the display. The CNR

rneasures the relationship between the signal difference versus the noise levels.

However, the signal difference and noise levels are changed independently after

subtraction, and the range of signal levels is reduced, so CNR before and after

subtractions cannot be compared directly. In short, if we know the gray scale

level of one signal, we can deduce the gray scale level of the other signal from

contrast sensitivity, but cannot deduce it from the CNR.

The phantom studies and patient studies indicate that MIP subtraction can

significantly enhance arteriai conspicuity, but at the expense of the Ioss of SNR.

2.5.2 Partial Volume Effects

As we introduced in section 1.2.2.7, subtracting complex numbers is preferred to

recover vessel signals particularly in thick slab projection 2D imaging. In 3D

imaging as performed here, the image resolution is generally high enough to

avoid partial volume effects in major vessels. However, for some small vessels

and the edges of vesseis, partial volume effects do still exist. But even if

subtracting cornplex numbers can recover part of the vessel signals, these

signals have to be compared with the background tissue signals along the

projection ray when doing MIP. As a result, in many cases, the vessel signals

retained by subtracting complex nurnbers will be lost again after MIP. On the


contrary, in MIP subtraction, the increase of the background mean associated

with the MIP operation will be subtracted out; therefore, the background level can

be reduced below that of complex subtraction and magnitude subtraction. Thus

some small vessel signais can be seen after MIP subtraction which are not seen

with the other approaches (Figure 2.1, 2.8). In 3D imaging, the change of

background level by MIP generally is more significant than partial volume effects.

2.5.3 Subtraction in the Stepping-Table Protocol

In a single-injectionktepping-table protocol (section 1.2.2.5), the concems of

subtraction are different from those discussed here. The scan is perfonned

mostly during the arterial phase of the contrast agent. Generally venous

enhancement is not a great concem. The purpose of subtraction is to subtract out

the background tissues and thus increase arterial conspicuity. In the mask data

set, there is no contrast enhancement and vessel signal intensities are lower than

surrounding tissues. As a result, in the MIP image of the mask data set, the

vessel signals will be replaced by higher intensity tissue signals on the projection

rays. This makes MIP subtraction less meaningful, because it will subtract higher

signals in regions corresponding to arteries. However, we should also consider

the increase of background tissue levei in complex subtraction and magnitude

subtraction. The relative performance of the various algorithms depends on

whether the decrease of arterial signal or the increase of background level is

more significant.


This situation can be demonstrated by the lower leg scan in our clinical studies,

which actually can be considered as a single-injection scan by itself. In the mask

MIP image in Figure 2.10, the arterial signals were completely replaced by the

greater leg tissue signals. However, comparing the result of three subtracted

images, MIP subtraction still generates the best arterial conspicuity. The statistics

in Table 2.8 also demonstrate that the best background suppression occurs after

MIP subtraction. In Table 2.9, the arterial-tissue conspicuity is improved more

significantly after MIP subtraction compared with complex and magnitude

subtractions. This result suggests that in a single-injection/stepping-table

protocol, the increased background signal level due to MIP should also be a

major consideration.

Figure 2.10 Subtraction in lower legs. a. mask MIP image; b. contrast-enhanced MIP image before subtraction; c. MIP image after complex subtraction; d. MIP image after magnitude subtraction; e. after MIP subtraction.


Table 2.8 Statistics of Background Tissues in the Lower Leg

Table 2.9 Conspicuity and SNR Statistics in the Lower Leg

Tissue Mean I I I -

Tissue Std Dev

2.5.4 Other Volume Rendering Algorithms

Complex Sub 1

0.1 3 1 O. 14 1 0.1 7

Artery-Tissue Conspicuity Arterial SNR

Of course, if we can develop another volume rendering algorithm rather than

MIP, in which the vessel signals are not compared with the maximum

background tissue signal, then subtracting complex numbers will be the ideal

choice in theory. However, as we showed in section 1.2.2.7, other volume

rendering algorithms al1 involve some kind of segmentation step, and a certain

threshold has to be applied. Generally the threshold has to be chosen by the

operator either in advance or interactively, so the objectivity of the algorithrns will

be in question. Furtherrnore, the threshold is chosen based on noise or

background statistics, so some low intensity small vessels generally will be lost.

These two factors reduced the popularity of these alternative volume rendering

algorithms.

In X-ray digital subtraction angiography, two sum projection images are

subtracted. Sum projection keeps the vessel width information along the

projection ray, while MIP projection only shows the brightest single pixel, no

matter how thick the vessel is along the projection ray. Thus, MIP processing will

Mag Sub 0.86

Complex Sub 0.57 7.86

MIP Sub 0.28

Mag Sub 0.63 7.86

MIP Sub 0.88 7.86

Before Sub 0.52 15.69


cause under-estimation of a stenosis. To overcome this problern, multi-angle MIP

is definitely necessary in clinical practice. While rotating the 3D matrix, the

projection length may increase to 256 or 51 2 pixels, so the background level is

not constant. This should be kept in mind if the vessels appear to change

diameter during rotation.

2.5.5 Artifacts by Subtraction

Several imaging factors will influence the effect of the subtraction algorithms. The

most important concem is the timing of the contrast agent. If the acquisition starts

before the contrast has filled the complete arterial vasculature in the FOV, sorne

arterial information will be lost. In a multiple-injection protocol, the steady state

information of these arteries still can be seen in the non-subtracted image. This is

helpful to distinguish timing artifacts from arterial occlusion. However, after

subtraction, the steady state information is no longer available. Furthermore, this

may cause the loss of signal after MIP subtraction such as in Figure 2.9. On the

other hand, if the acquisition starts too late, the contrast agent flows into the

veins during the acquisition of the center of k-space and some venous signais wili

also be enhanced. This is a potential problem of the distal region in stepping

table protocols. Subtraction will not only increase the conspicuity of arteries, but

also will increase the conspicuity of veins in this case. Therefore, a good timing

method (automatic trigger, section 1.2.2.4) is a prerequisite to the success of

subtraction.


Another concem is patient motion. This can cause mis-registration during

subtraction and introduce unpredictable artifacts. To avoid this probfem, generally

the patient's feet are fastened together during the scan. In 3D centric ordered

acquisition, as long as the motion does not occur at the start (the acquisition of

central k-space), the motion artifact will be averaged over the entire 3D data set

and may not significantly influence the image quality8. In our five clinical studies,

we did not observe significant mis-registration problems.

For multiple-injection protocols, a particular concem is the potential change of the

contrast concentration between the mask scan and the corresponding contrast-

enhanced scan. In our protocol, we start the mask scan after a 20 localizer scan

to prescribe the 3D volume of interest, which takes about 3 minutes. The fresh

contrast agent was then injected and timed for the contrast-enhanced scan,

which takes about 4 minutes. We thus can complete al1 upper leg scans in 7

minutes. Generally, Magnevist in the body has a half-life of about 30 minutes

after becoming uniformly distributed. Thus its concentration should not drop

significantly in the 7 minutes. In our experience, we did not observe much change

in the signal intensities. However, if the concentration does drop significantly,

there will be negative values in veins during magnitude subtraction and MIP

subtraction. In magnitude subtraction, these negative veins will be replaced by

surrounding tissues along the projection ray after MIP, so will not cause any

problem. In MIP subtraction, these negative values will raise the base gray scale

level of the image if we display the full signal range. Actually, because of the

variance of the signals, even without the concentration change, there are also


negative values after subtracting the two MIP images. We tried both thresholding

these negative values to zero and getting their absolute values. The two

approaches yiefded similar results; therefore, in our study, the negative values

are within the range of the variance of background tissue level and do not

interfere with the depiction of arteries.

2-5-6 Do We Need Subtraction in 30 Scans?

It has been demonstrated that, in the lower extremities, su btraction techniques

work better than fat suppression techniques to increase arterial conspicuiv5.

However, a critical question is how much subtraction improves conspicuity? In

Table 2.5 and 2.9, we can see the artery-vein conspicuity does improve after

subtractions, and artery-tissue conspicuity can be either improved or reduced

after complex and magnitude subtractions, depending on the initial contrast of the

data set. The arterial conspicuity always improves after MIP subtraction, and to a

greater extent compared with the other two algorithms. However, the expense is

the significantly lower arterial SNR after al1 subtractions. Furthemore, the

potential artifacts of subtractions are also considerations. In this case, the

improvement of arterial conspicuity after MIP subtraction is the major attractive

feature in subtractions.


2.6 Conclusion

Image subtraction iç the most popular rnethod for increasing arterial conspicuity

in ciinical first-pass contrast-enhanced MR angiography. 30 imaging has proved

to be superior to 2D thick slab projection protocol. Because of the MIP

processing involved in 3D display, cornplex subtraction does not yield the best

arterial conspicuity in most clinical cases if the resolution is high enough to avoid

partial volume effects in most vessels. MIP subtraction generally gets the best

background tissue suppression and arterial conspicuity, when the contrast agent

yield higher arterial signals than the surrounding tissues in the raw data sets.

Therefore, in multiple-injection protocols, when image spatial resolution is high

enough to avoid partial volume effects in rnost arteries, and contrast

concentration is high enough to enhance artenal signals over surrounding tissues

(both of which are generally satisfied), MIP subtraction should be the first choice

for a processing method to elirninate venous signals. In the lower leg station,

because no venous contamination existed, subtraction can be omitted. In single-

injection/stepping-table protocols, because of the lirnited irnaging time, spatial

resolution generally is not sufficient to avoid partial volume effects. Complex

subtraction is still needed to recover the resulting signal loss. The un-subtracted

data set, particularly the source slice images, also should be evaluated to correct

potential artifacts.

Chapter 3

Future Directions

Subtraction has been demonstrated to be an effective technique to enhance

arterial conspicuity in first-pass state MR angiography. However, due to the

limited time window of the first-pass state (which is about one minute for the

lower extrernities), image resolution has to be compromised. Once the

gadolinium-based contrast agents like Gd-DTPA enter the capillary bed, they will

diffuse through the endothelial wall of the capillaries, and equilibrate with

interstitial fluid in a time on the order of minutes due to their srnall molecular

weight and non-binding nature. it is estimated that 50% of Gd-DTPA is cleared

from the vascular space into the extravascular compartment on the initial pass

through the cap il la rie^^^. The contrast concentration in arteries is therefore

significantly reduced after the first pass. Furthemore, the contrast agent leaking

into the surrounding tissues will increase the signal intensities of the background

and decrease the arterial conspicuity. Although the automatic triggering and

centric order acquisition techniques help to acquire more arterial information

during the first pass of the contrast agents, the uniformity of the contrast

distribution in the arteries varies among patients. Some small arteries are likely to

be lost in the images due to a delayed enhancement. These considerations lead

to the development of blood pool contrast agents which remain in the vessels for

53

Chapter 3. Future Directions 54

a much longer tirne than the extracellular contrast agents. The scan of blood pool

contrast agents in steady-state may help to achieve a higher image resolution

and arterial conspicuity, especially in diseased regions. However, venous signals

will also be enhanced in steady-state, which interfere the depiction of arterial

vasculature in MIP images. Subtraction techniques cannot be applied in steady-

state imaging due to the equilibrium nature. Other segmentation techniques are

required to separate arteries from veins. In the following, we wili discuss the

development of the blood pool contrast agents and several promising

segmentation approaches.

3.1 Blood Pool Contrast Agents

Currently there are two promising blood pool contrast agents undergoing clinicai

trials: NC10015O lnjection and MS-325. Using different rnechanisms, they

achieve long intravascular half-lives. We discuss their specific properties

separately.

3.1.1 NC100150 lnjection

NC100150 lnjection (Clariscan, Nycomed Amersham, Oslo, Norway) is an

ultrasmall superparamagnetic iron oxide (USPIO) blood pool agent with a single

iron oxide crystal core of 5-7nrn diameter, stabilized with a carbonhydrate-

polyethylene glycol coat binder, which results in a total particle diameter of

20nmS8. lnjection at a dose of 5mg Fe/kg reduces blood Tl to below 100 msec


for greater than 2 hoursS9. Ultimately, the iron is taken up by macrophages and

rnetabolized by the liveP8.

A major concern with USPIO contrast agents is their T2' shortening effects. The

r l relaxivities of Gd-based contrast agents are only slightly srnaller than their r2

relaxi~i t ies~~. However, the r2hl value of NC100150 Injection is about 1.8 at

0.5T, 37OC, and about 2.5 at 1.5T. Therefore, TE must be rninimized to rninimize

signal loçs from the T2* shortening effects.

Significant SNR and CNR increases for NC100150 lnjection compared with Gd-

based extracellular contrast agents were achieved in clinical trials with TI

weighted sequen~es~~? The blood T l relaxation tirne measured 60 to 90

minutes after the intravenous injection does not significantly differ from that

obtained immediately after the injection. 0.75~0.75~0.75rnm image spatial

resolution can be achieved for the lower leg station without the time limitation of

extracellular agents63. The accuracy of measuring stenoses is also significantly

better than that with extracellular contrast agents.

MS-325 (Angiomark, Epix Medical, Cambridge, MA) is a small molecule Gd-

chelate agent binding strongly and reversibly to human serum albumin (about

96%), creating a macromolecular complex with a blood pool d i s t r i b u t i ~ n ~ ~ * ~ ~ . The

equilibrium between free and protein-bound MS-325 is such that a small amount


of unbound MS-325 is always present, ensuring efficient renal excretion. This

unique mechanisrn of action overcornes the tissue retention problerns

characteristic of earlier blood pool prototypes containing gadolinium chelates

irreversibly bound to large polymeric structure^^^.

The r l relaxivity of MS-325 in human plasma is about 6 to 10 times that of Gd-

DTPA. The T2' shortening effects are substantially smaller compared with

USPIO agents, yielding the potential for higher SNR and CNR~? A

0.05mmoVkg dose is generally applied in clinical t r i a ~ s ~ ~ - ~ ~ . Although in theory

higher doses rnay achieve better SNR, the binding sites on alburnin may becorne

saturated with MS-325. This results in a lower binding percentage and fast

elimination of the unbound fraction of MS-3~5'~. Generally MS-325 can provide

sufficient intravascular contrast for up to one hour6? MS-325 is not metabolized

in the body and is excreted completely by the kidneys with an elimination half-life

about 2-3 hours.

Although the first-pasç applications of blood pool contrast agents showed higher

SNR and CNR over those with extracellular agents, the more attractive feature is

certainly the potentially higher spatial resolution in steady state irnaging.

However, a big problern with the blood pool contrast agents in the steady state is

the venous enhancement. The overlay of arterial and venous signals in the MIP

images makes distinguishing arteries from veins di f f i~ul t~ ' -~? Subtraction

techniques cannot be applied here because we use the steady state information

rather than the first-pass information. Therefore, other segmentation techniques

Chapter 3. Future Directions

must be applied to separate artenes from veins before doing the MIPs. Various

approaches have been tried but the results are still not sa t i s fa~ to ry~~-~~ . Before

the invention of a successful segmentation technique, it will be hard to apply the

steady state blood pool contrast agents in clinical practice.

3.2 Artery-Vein Segmentation

Long before the invention of blood pool contrast agents, vessel segmentation

techniques have been investigated to separate vessels from background tissues,

39-49 mostly to avoid the background increment problem of MIP processing .

However, due to their tendency to lose small vessels and due to the complexity

of the algorithms, these techniques are seldom applied in clinical practice. Whole

volume MIP is still the most widely used algorithm in 3D MR angiography. This

wams us that for clinical applications, the artery-vein segmentation technique has

to be as simple as possible and as robust as possible. Several protocols have

been proposed.

3.2.1 Connectivity-based Algorithms

Connectivity-based algorithms (also called seed-growing algorithms) are the

classic techniques in vessel segmentation3949. After the selection of a seed pixel

manualiy in a targeted vessel, the signal intensities of its irnmediately

surrounding pixels are checked and the pixels within a pre-set variance are

judged as the vessel signals. The same algorithm is applied iteratively for the


new vessel pixels untii no more pixels can be classtied as vessel pixels. In

theory, this algorithm can segment a vascular tree from background signals if a

proper threshold is prescribed. If the seed pixel is intentionally selected in a

venous branch, the venous vasculature then can be segmented out. However,

the effectiveness of the connectivity-based algorithms varies arnong individual

cases. The threshold has to be set according to the signal statistics of the

specific case, and the selection of the seed pixel also influences the result. If

there are several non-connected branches (e.g. disconnected by the limitation of

the FOV), multiple seed pixels have to be set. The involvement of too many

interactive interventions impedes the wide application of the approach.

Particularly, where the arteries and veins grow next to each other, the

connectivity-based algorithm may cause mis-classification. It has been well

recognized that pure image post-processing techniques are hard to be applied in

medical imaging because of the variability of pathological condition^'^.

3.2.2 Temporal Correlation-based Techniques

The first-pass information can be used for the artery-vein segmentation in the

steady state. It has been proposed that a series of high temporal resolution, low

spatial resolution 3D data sets be acquired during the first pass of the contrast

agent for use in ~egmentation~'-'~. High spatial resolution acquisition then can be

perfoned in steady state. The time-resolved signal intensity curves of a selected

arterial sample pixel and a venous sample pixel then can be detemined from the

series of first-pass data sets. Due to the delayed enhancement of venous signals,


there will be a significant time shift between the two curves. Correlations of the

two curves and the signal intensity cutve of each pixel then can be done and

thresholds can be set for the correlation coefficients to classify the arterial pixels

and the venous pixels as well as the background pixels. The sarne portion of k-

space data for the steady state acquisition then can be reconstructed and the

corresponding venous pixels and background pixels can be segmented

according to the first-pass data sets. The high spatial resolution part of k-space

data is also reconstructed and added to the segmented low resolution data.

Although the high-resolution venous signals still exist, because there are no low-

resolution venous signals, the corresponding veins are generally not visible

structures and do not influence the depiction of arterial signals.

The effectiveness of this technique depends on the correlation of the sarnple

pixels and the other pixels. The correlation depends on the unifom distribution of

the contrast agent. If there is a significant enhancernent delay in the stenosed

arterial region, the correlation segmentation may fail. Furthemore, although the

high spatial resolution venous signals do not f o m visual structures, they will

appear as high signal intensity background noise, which also may be a concem.

Finally, the possibility of mis-registration between the first-pass data and the

steady state data is also a concem.

Despite the potential questions, the temporal correlation-based techniques

combine the dynamic information in first pass and the high spatial resolution


information in steady state. They will be further evaluated and developed in the

future.

3.2.3 Phase-Contrast-based Techniques

The main arteries and veins in the lower extremities run longitudinaily and the

blood flows are in opposite directions. Therefore, arteries and veins will induce

different phase information in the presence of a velocity-encoded sequence

(section 1.2.1 -2). The phase difference image between a flow-compensated data

set and a non-flow-compensated data set can be used as the flow encoding of

73-74 the arterîal and venous signals . However, this technique only works for the

longitudinal vessels and is not sensitive in the in-plane direction. A turbulent flow

region also will introduce phase artifacts. The primary advantage of contrast-

enhanced MR angiography is its flow-independent nature. It is therefore not

reasonable using a flow-dependent technique alone to guide segmentation.

3.2.4 Oxygen Level-based Techniques

It has been shown that T2 measurernent can be used to determine the blood

oxygen level measurement in vivo7? The relationship between T2 and the

oxygen level is:

I 1

T 2 , T2,

where %Hb02 = blood hemoglobin oxygen saturation,

T2, = T2 of oxygenated blood.


For absolute oxygen level measurernent, k needs to be calibrated on an in vitro

blood sample T2 measurement so that the in vivo relationship behnreen T2 and

oxygen level can be established. However, to separate arteries and veins, we

only need to know the relative oxygen levels. For this case, the k can be

canceled by introducing a reference arterial pixel:

If we select an arterial pixel and suppose its oxygen level to be e.g. 97%, then we

can get the relative oxygen levels of other pixels. The significant oxygen level

difference between arteries and veins can be useful for segmentation purposes.

To our knowledge, this technique has not been investigated in detail in

conjunction with blood pool agents.

3.3 Summary

The newly developed blood pool contrast agents can provide significantly higher

SNR and CNR than extracellular agents. Furthemore, they provide a much wider

time window for high spatial resolution imaging. Due to the venous contamination

problem, artery-vein segmentation techniques will be a critical factor in the

successful application of blood pool contrast agents.

References

Branchereau A. Critical limb ischemia. Futura Pub Co., New York, 1999

Ouriel K. Lower extremity vascular disease. W B Saunders Co., Philadelphia, 1995

Maritini F. Fundamentals of anatomy and physiology. Prentice Hall, New Jersey, 1992

Stary HC. Atlas of atherosclerosis: progression and regression. Parthenon Pub Group, New York, 1999

Kappert A. Diagnosis of Peripheral Vascular Disease. Bem H. Huber, 1971

Kim D and Orron DE. Peripheral Vascular lmaging and Intervention. Mosby- Year Book, St. Louis, 1992

Waugh JR and Sacharias N. Arterialgraphic complications in the DSA era. Radiology 182:243-246, 1992

Prince MR, Grisi TM and Debatin JF. 3D contrast MR angiography. Springer- Verlag, Berlin, 1 999.

Sueyoshi E, Sakamoto 1, Matsuoka Y, Ogawa Y, Hayashi H, Hashmi R and Hayashi K. Aortoiiiac and lower extremity arteries: comparison of three- dimensional dynamic contrast-enhanced subtraction MR angiography and conventional angiography. Radiology 21 0:683-688, 1999

10.Quinn SF, Sheley RC, Semonsen KG, Leonardo CJ, Kojima K and Szumowski J. Aortic and lower-extremity arterial disease: evaluation with MR angiography versus conventional angiography. Radiology 206:693-701, 1 998

11.Thumher SA, Dorffner R, Thumher MM, Winkelbauer FW, Kretschmer G, Polterauer P and Lammer J. Evaluation of abdominal aortic aneurysrn for stent-graft placement: comparison of gadolinium-enhanced MR angiography versus helical CT angiography and digital subtraction angiography. Radiology 2051341 -352, 1997

12.Link J, Steffens J, Brossmann J, Graessner J, Hackethal S and Heller M. lliofemoral artenal occlusive disease: contrast-enhanced MR angiography for

References

preinterventional evaluation and follow-up after stent placement. Radiology 21 2~371-377, 1999

13. Potchen EJ, Haacke EM, Siebert JE and Gottschalk A. Magnetic resonance angiography: concepts & applications. Mosby, St. Louis, 1993

14. Stark DD and Bradley WG. Magnetic resonance irnaging. Mosby-Year Book, St. Louis, 1992

15. Felix R, Heshiki A, Hosten N and Hricak H. Magnevist monograph. Blackwell Science, Berlin. 1997

16. Crawley AP, Wood ML, HenkeIman RM. Elimination of transverse coherence in FLASH MRI. MRM 8:248-260. 1988

17.Zur Y, Wood ML and Neuringer LJ. Motion-insensitive, steady-state free precession imaging. MRM 16:444-459. 1990

18.Bampton A, Riederer SJ and Korin HW. Centric phase-encoding order in three-dimensional MR-RAGE sequences: application to abdominal imaging. JMRI 2:327-334, 1992

19.Wilrnan AH and Riederer SJ. lrnproved centric phase encoding orders for th ree-d imensional magnetization-prepared M R angiography. M RM 36384- 392, 1996

20. WiIman AH and Riederer SJ. Performance of an elliptical centric view order for signal enhancement and motion artifact suppression in breath-hold three- dimensional gradient echo imaging. MRM 38:793-802, 1997

21. Kim JK, Farb RI and Wright GA. Test bolus examination in the carotid artery at dynamic gadolinium-enhanced MR angiography. Radiology 206:283-289, 1998

22. Wilrnan AH, Riederer SJ, King BF, Debbins JP, Rossrnan PJ and Ehman RL. F!uoroscopically triggered contrast-enhanced three-dimensional MR angiography with elliptical centric view order: application to the renal arteries. Radiology 205:137-146, 1 997

23.WiIman AH, Riederer SJ, Huston III J, Wald JT and Debbins JP. Arterial phase carotid and vertebral artery imaging in 3D contrast-enhanced MR angiography by combining fluoroscopic triggering with an elliptical centric acquisition order. MRM 40:24-35, 1998

References

24. Foo T, Saranathan M, Prince MR and Chenevert TL. Automated detection of bolus arrivai and initiation of data acquisition in fast, three dimensional, gadolinium-enhanced MR angiography. Radiology 203:275-280, 1997

25. Ho KY, Leiner T, de Haan MW, Kessels A, Kitslaar P and van Engelshoven J. Peripheral vascular tree stenoses: evaluation with moving-bed infusion- tracking MR angiography. Radiology 206:683-692, 1 998

26.Ho KY, Leiner T, de Haan MW, Kessels A, Kouwenhoven M and van Engelshoven J. lmage interpretability of penpheral MR angiography using the MoBI-track technique. ISMRM 99:350, 1999

27.Boudewijn G, Vasbinder Cl Kaandorp DW, Ho KY, de Haan MW and van Engelshoven J. Gadolium-enhanced 3 0 renal MRA: paired comparison of bolus-timing-independent fast multiphase acquisition with standard bolus- timed 3D MRA in 20 subjects. ISMRM 2000:695,2000

28.Schoenberg SO, Londy F, Licato P, Williams D, Wakefield T and Chenevert TL. Multiphase-multistep 3D gadolinium-enhanced MRA of the abdominal aorta and run-off vessels. ISMRM 2000:690,2000

29. Davis LR, Duerinckx AJ and Shaaban A. Multistation gadolinium-enhanced MR angiography of the lower extremities: effect of multiple injections on image quality. ISMRM 99:1224, 1999

30. Lee JM, Chang Y, Kim YJ and Kang DS. The effectiveness of digital subtraction technique in contrast-enhanced M R angiography. ISMRM 99:1233, 1999

31 .Ruehm SG, Pfammatter Tl Schmidt M, Schneider E and Debatin JF. 3D contrast-enhanced MRA of the runoff vessels: value of image subtraction. ISMRM 99:352, 1999

32. Leung DA, Pelkonen Pl Hany TF, Zimmermann G, Pfammatter T and Debatin JF. Value of image subtraction in 3D gadolirnium-enhanced MR angiography of the renal arteries. JMRl 8:598-602, 1998

33.Ho KY, de Haan MW, Kessels A, Kitslaar P and van Engelshoven J. Peripheral vascular tree stenoses: detection with subtracted and nonsubtracted MR angiography. Radiology 206:673-681, 1 998

34.Leiner T, Ho KY and van Engelshoven J. Image subtraction in contrast- enhanced peripheral MRA. ISMRM 2000:1808,2000

References

35.Gaucher H, Lefevre F, Lehalle B, Argaud C and Regent D. Aorta and lower extremities bolus chasing MRA: evaluation of background signal reduction techniques. ISMRM 2000:184,2000

36.Wang Y, Johnston DL, Breen JF, Huston III J, Jack CR, Julsnid PR, Kiely MJ, King BF, Riederer SL and Ehrnan RL. Dynarnic MR digital subtraction angiography using contrast enhancernent, fast data acquisition, and cornplex subtraction. MRM 36:551 -556, 1996

37. Brown DG and Riederer SJ. Contrast-to-noise ratios in maximum intensity projection images. MRM 23:130-137, 1992

38.Sun Y and Parker DL. Performance analysis of maximum intensity projection algorithm for display of MRA images. IEEE trans on Medical lmaging 18(12):1154-1169, 1999

39.Lin W, Haacke EM, Masaryk TJ and Smith AS. Automated local maximurn- intensity projection with three-dimensional vesse1 tracking. JMRl 2:519-526, 1992

40.Korosec FR, Weber DM, Mistretta CA, Turski PA and Bernstein MA. A data adaptive reprojection technique for MR angiography. MRM 24:262-274, 1992

41 .Cline HE, Thedens DR, Meyer CH, Nishirnura DG, Foo TK and Ludke S. MR coronary angiography using connectivity. ISMRM 99: 1 251 , 1 999

42.Cline HE, Durnoulin CL, Lorensen WE, Souza SP, Adams WJ. Volume rendering and connectivity algorithms for MR angiography. MRM 1 8:384-394, 1991

43.Cline HE, Lorensen WEI Souza SP, Jolesz FA, Kikinis R, Gerig G and Kennedy TE. 3D surface rendered MR images of the brain and its vasculature. J Computer Assisted Tomography 1 5(2):344-351, 1991

44. Cline HE, Lorensen WE, Kikinis R and Jolesz F. Three-dimensional segmentation of MR images of the head using probability and connectivity. J Computer Assisted Tomography 14(6):1037-1045, 1990

45.Cline HE, Lorensen WE and Ludke S. Two algorithms for the three- dimensional reconstruction of tomograms. Med. Phys. 15(3):320-327, 1988

46.Cline HE, Dumoulin CL, Hart HR, Lorensen WE and Ludke S. 3D reconstruction of the brain frorn magnetic resonance images using a connectivity algorithm. MRI 5:345-352, 1987

References

47.Hu X, Alperin N, Levin DN, Tan KK and Mengeot M. Visulization of MR angiographic data with segmentation and volume-rendering techniques. JMRl 1 (5):539-546, 1 99 1

48. Lin W, Haacke EM, Smith AS and Clampitt ME. Gadolinium-enhanced high- resolution MR angiography with adaptive vesse1 tracking: preliminary results in the intracranial circulation. JMRl 2(3):277-284, 1992

49.Saloner D, Hanson WA, Tsuruda JS, van Tyen R, Anderson CM and Lee RE. Application of a connected-voxel algorithm to MR angiographic data. JMRl 1 1423-430, 1 991

50.Kruger RA and Riederer SJ. Basic concepts of digital subtraction angiography. G. K. Hall Medical Publishers, Boston. 1984

51. Henkelman RM. Measurement of signal intensities in the presence of noise in MR images. Med. Phys. 12(2):232-233,l985, 13(4):544, 1986

52. Edelstein WA, BottomIey PA and Pfeifer LM. A signal-to-noise calibration procedure for NMR imaging system. Med. Phys. 1 1 (2):180-185, 1984

53.Firbank MJ, Coulthard A, Harrison RM and Williams ED. A comparison of two methods for measuring the signal to noise ratio on MR images. Phys. Med. Biol. 44:261-264, 1 999

54.Sijbers J, Scheunders P, Bonnet N, van Dyck D and Raman E. Quantification and improvement of the signal-to-noise ratio in a magnetic resonance image acquisition procedure. MRI l4(lû):l l57-l l63, 1996

55.Sijbers J, den Dekker AJ, van Audekerke J, Verhoye M and van Dyck D. Estimation of the noise in magnitude MR images. MRI 16(1):87-90, 1998

56. Gudbjaitsson H and Patz S. The Rician distribution of noisy MRI data. MRM 34191 0-91 4, 1 995

57.Sijbers J, den Dekker AJ, Scheundes P and van Dyck D. Maximum-likelihood estimation of Rician distribution parameters. IEEE trans on Medical lmaging 17(3):357-361, 1998

58.Kroft L and de Roos A. Blood pool contrast agents for cardiovascular MR imaging. JMRl 10:395-403, 1999

59.Taylor AM, Panting JR, Keegan J, Gatehouse PD, Amin D, Jhooti P, Yang GZ, McGill S, Burman ED, Francis JM, Firmin DN and Pennell DJ. Safety and preliminary findings with the intravascular contrast agent NC100150 injection for MR coronary angiography. JMRl 9:220-227, 1 999

References

6O.Anzai Y, Prince MR, Chenevert TL, Maki JH, Londy F, London M and McLachlan SJ. MR angiography with an ultrasmall superparamagnetic iron oxide blood pool agent. JMRl7:209-214, 1997

61. Niessen WJ, van Swijndregt Ml Elsman B, Wink O, Viergever MA and Mali W. Steady state and first pass blood pool agent contrast enhanced MRA. ISMRM 2000:467,2000

62. Niessen WJ, van Swijndregt M, Elsman B, Wink O, Viergever MA and Mali W. Blood pool MR angiography of the peripheral vasculature. ISMRM 2000:1232, 2000

63.Ho KY, Leiner Tl de Haan MW, Kessel A and van Engelshoven J. Cornparison between blood pool-enhanced and gadolinium-chelate-enhanced MR angiography for imaging the peripheral arteries. ISMRM 99:12301 1 999

64Schmidt M, Weishaupt Dl Ruehm SG, Binkert C, Patak MA, Steybe F and Debatin JF. MRA of the aorta, renal and pelvic arteries: extracellular vs. intravascular contrast agents. ISMRM 99:356, 1 999

65.Knopp MV, von Tengg-Kobligk Hl Floemer F and Schoenberg SO. Contrast agents for M M : future directions. JMRl 1 O:314-316, 1999

66.Lauffer RB, Parmelee DJ, Dunham SU, Ouellet HS, Dolan RP, Witte S, McMurry TJ and Walovitch RC. MS-325: Albumin-targeted contrast agent for MR angiography. Radiology 207:529-538, 1998

67. Grist TM, Korosec FR, Peters DG, Witte S, Walovitch RC, Dolan RP, Bridson WE, Yucel EK and Mistretta CA. Steady-state and dynamic MR angiography with MS-325: initial experience in hurnans. Radiology 207539-544, 1998

68.Bock M, Schoenberg SO, Flomer F and Schad LR. Separation of arteries and veins in 3D MR angiography using correlation analysis. MRM 43:481-487, 2000

69.Bock M, Schoenberg SO, Flomer F, Grau A, Strecker R and Schad LR. Artery-vein separation in 3D contrast-enhanced pulmonary MRA using correiation analysis. ISMRM 99:486, 1999

70.Du J, Mazaheri Y, Carroll TJ, Esparza Coss El Grist TM and Mistretta CA. Automated Region-specific WRAC segmentation in peripheral MRA. ISMRM 2000:1807,2000

71.Mazaheri Y, Carroll TJ, Korosec FR, Block W, Du J, Grist TM and Mistretta CA. High resolution CE-MRA using dual-resolution acquisition and

References

segmentation based on spatial frequency-dependent 2D temporal correlation. ISMRM 2000:189,2000

72. Mazaheri Y, Carroll TJ, Mistretta CA, Korosec FR and Grist TM. Vesse1 segmentation in 3D MR angiography using time resolved acquisition curves. ISMRM 99:218l, 1999

73.Svensson J, Leander P, Stahlberg F, Thomsen C, Sjoberg SI Olsson LE. Segmentation of arteries from veins in contrast-enhanced 3D magnetic resonance angiography of the lower extremities. ISMRM 2000:1809,2000

74.Bluemke DA, Darrow RD, Gupta R, Tadikonda SK and Domoulin CL. 3D contrast enhanced phase contrast angiography: utility for arteriaVvenous segmentation. ISMRM 99: 1 237, 1999

75. Wright GA, Hu BS and Macovski A. Estimation oxygen saturation of blood in vivo with MR imaging at 1.5T. JMRl 1 :275-283, 1 991

Documents

Analysis of Subtraction Methods Contrast …...Abstract Analysis of Subtraction Methods in 3D Contrast-Enhanced MR Digital Subtraction Angiography Yuexi Huang Master of Science 2001