Separation of arteries and veins using flow-induced phase effects in contrast-enhanced MRA of the...

Separation of arteries and veins using flow-induced phase effects incontrast-enhanced MRA of the lower extremities

Jonas Svenssona, Peter Leanderb, Jeffrey H. Makic, Freddy Stahlbergd, Lars E. Olssona,*aDepartment of Radiation Physics, Malmo University Hospital, SE-205 02, Malmo, Sweden

bDepartment of Diagnostic Radiology, Malmo University Hospital, SE-205 02, Malmo, SwedencSeattle VAHCS, Department of Radiology (114), 1660 S. Columbian Way, Seattle, WA 98108, USA

dDepartment of Radiation Physics, Lund University Hospital, SE-221 85 Lund, Sweden

Received 2 November 2001; accepted 26 January 2002

Abstract

In 3-D contrast-enhanced magnetic resonance (MR) angiography of the lower extremities the goal is most often to enhance arterialstructures while keeping veins and surrounding tissue unenhanced. Imaging during steady-state concentration of a blood pool agent or duringpoor timing of an extra-cellular contrast medium may result in simultaneous venous enhancement, making interpretation of the angiogramdifficult.

The aim of this study was to develop a post-processing method to separate the arteries from the veins in standard contrast-enhanced MRangiograms.

The method was based on the different accumulation of flow-induced phase in the arteries and veins of the lower extremities. The methodwas tested in both phantom experiments and volunteers undergoing 3-D contrast-enhanced MR angiography using both an extra-cellularcontrast medium and a blood pool agent.

In the phantom studies, opposite directional flow was successfully separated at mean flow velocities as low as 9 cm/s. In the volunteerstudies, the larger veins were successfully extinguished while the larger arteries were left unaffected. In smaller vessels with low flowvelocities the separation was less successful. This was most apparent in vessels not oriented superior-inferior.

The method developed here is promising for separating arteries from veins in contrast-enhanced MR angiography although the resultscould be further improved by either a different pulse sequence design or combining this method with other segmentation methods. © 2002Elsevier Science Inc. All rights reserved.

Keywords: Contrast-enhanced; Separation; Artery; Vein; MR; Angiography

1. Introduction

3-D Contrast-enhanced (CE) magnetic resonance an-giography (MRA) of large regions of the peripheral vascu-lature using a conventional extra-cellular contrast medium(CM) is complicated. It is commonly performed with care-ful injection timing and a “bolus chase” technique withshort acquisition times [1–5]. A new contrast medium typecalled blood pool agents (BPA), with long vascular half-lifecompared to imaging time [6–11] could considerably facil-itate imaging of the peripheral vessels. A BPA eliminatesthe need for complicated timing procedures and instead

permits longer acquisition times during the equilibriumphase of CM concentration. This longer acquisition timecan be used to increase the overall image quality withtechniques such as increasing the spatial resolution and/orthe number of signal averages [6,9,12]. If MRA imaging isperformed during CM concentration equilibrium however,new problems arise. The concentration of CM will be ap-proximately the same in all vessels, leading to similar signalenhancement in both arteries and veins and subsequentdifficulties in separating them in the angiogram [6,7,12].The same problem can occur if extra-cellular CM is usedand the injection timing is poor [13–15].

Consequently, in arterial imaging, there is a need for apost-processing technique in order to extinguish the veins inthe acquired image volume. Several methods have beenproposed. One technique is based on the differences in

* Corresponding author. Tel.: �46-040-33-10-21; fax: �46-040-96-31-85.

E-mail address: [email protected] (L.E. Olsson).

Magnetic Resonance Imaging 20 (2002) 49–57

0730-725X/02/$ – see front matter © 2002 Elsevier Science Inc. All rights reserved.PII: S0730-725X(02)00479-4

blood oxygenation that lead to different susceptibility andconsequently to different phase signal in arteries and veins[16]. Another possible approach involves vessel trackingthrough seed growing and connectivity algorithms [17–19].Other methods utilize the differences in the temporal en-hancement patterns of arteries and veins following a CMbolus injection [20–23]. Most of the methods outlinedabove require specialized acquisition protocols. In order toperform segmentation using vessel tracking, the imagesmust be acquired with extremely high resolution given theclose proximity of the arteries and veins in the lower legs.If differences between arterial and venous temporal en-hancement patterns are utilized, this requires acquisition ofseveral 3-D volumes with a high temporal resolution duringthe arrival of the CM bolus.

Recently a method was proposed evaluating the possi-bility of using the flow-induced phase effects for distin-guishing arteries from veins [24,25]. Spins flowing in thedirection of a bipolar magnetic field gradient will accumu-late a phase angle proportional to the flow velocity. Thevessels in the legs are mainly oriented in the direction of theextremity, and the veins and arteries will therefore haveopposing flow direction. Hence if imaging is performedusing bipolar gradients in the superior-inferior direction, thephase of the MR-signal from the two vessel types willdiffer. This phase difference can be used to separate arteriesfrom veins [24,25].

The aim of this project is to evaluate flow-induced phaseeffects from pulse sequences used in lower extremity con-trast-enhanced MRA and develop a post-processing methodof arterio-venous segmentation based on these effects. Themethod will be demonstrated for suppression of venousenhancement in 3-D CE-MRA images using unfavorabletiming of an extra-cellular contrast medium, and in a pilotexperiment using a blood pool agent.

2. Materials and methods

2.1. Hypothesis

Spins moving in the direction of a spatially linear mag-netic field gradient accumulate a phase angle according toEq. (1) [26].

��t� � �t0

t

�d� � � �t0

t

G�� x��d� (1)

here � is the gyromagnetic ratio (2.68*108 rad/s/T for pro-tons), G(t) is the time dependent gradient strength (T/m)and x(t) is the time dependent position of the spins. Themost frequently used pulse sequences in CE-MRA are fastspoiled gradient echo sequences. The readout gradient inthese sequences is usually of bipolar form, and is designed

to set the integral in Eq. (1) to zero for stationary spins at thetime of the echo (TE). If the spins are not stationary, butinstead flow in the direction of the bipolar gradient, theaccumulated phase at TE will not be zero. For constant flowvelocity there is a linear relationship between phase at TEand flow velocity. This flow sensitivity may be increased byincreasing the gradient strength or by prolonging the dura-tion of the gradients. Instead of using a bipolar readoutgradient it is possible to design the readout gradient profileto yield zero phase for both stationary and flowing spins atTE. The sequence is then termed flow-compensated.

The arterial and venous flow in the legs is more or lessoppositely directed superior-inferior. If the readout is set inthis direction and flow compensation is not applied, theflowing arterial and venous blood results in positive andnegative phase accumulation respectively. This differencein phase values was recently suggested as a method forseparating the two vessel types in CE-MRA [24,25].

Calculations were performed using Eq. (1). G(t) was thesame as in the flow sensitive sequence used in the phantomand extra-vascular measurements below. The accumulatedphase at TE was calculated for the flow velocities used inthe phantom measurements.

2.2. Phantom measurements

A Magnetom Vision 1.5 T MR system (Siemens AG,Erlangen, Germany, software version Numaris 3, VB33D)and the extremity coil were used. A standard MRA fast lowangle shot (FLASH) pulse sequence (fi3d_2b195.ykc) wasmodified slightly to allow the use of flow-compensation inthe readout direction. This modification leads to a slightlyprolonged TE and somewhat higher flow sensitivity (whenused in not flow-compensated mode) as compared with theoriginal sequence. The imaging parameters used in all mea-surements (with or without flow-compensation) were iden-tical (TE/TR � 3.7/8.7 msec, FA � 30°, matrix � 512 �192 � 32, FOV � 320 � 240 � 64 mm, acquisition time �57 s, Bandwidth � 195 Hz/pixel).

The flow phantom consisted of two equally sized plasticpipes (diameter 7 mm) inside a plastic cylinder. The cylin-der was filled with GdCl3-doped water. The pipes wereinterconnected using a 2 m long soft PVC tube at one end ofthe phantom. The other ends of the pipes were connected totwo fluid containers with soft PVC tubes. Steady, non-pulsatile flow was established through the phantom using aheight difference between the two fluid containers. The flowwas oppositely directed through the two pipes. Flow veloc-ity was determined using the flow quantification packageavailable on the scanner. No CM was injected into thesystem during the experiments. The flow experiment fluidconsisted of distilled water doped to T1 � 100 msec usingGd3� (dissolved GdCl3). Imaging was performed at fivedifferent mean flow velocities: 0, 9, 15, 65 and 80 cm/s.Two image volumes were acquired for each velocity, one inflow-compensated mode and one without flow compensa-

50 Jonas Svensson et al. / Magnetic Resonance Imaging 20 (2002) 49–57

tion. Coronal plane acquisition was used with the readoutdirection parallel to the flow.

2.3. Volunteer measurement, conventional extra-cellularCM

CE-MRA using the conventional extra-cellular contrastmedium (Omniscan, Nycomed, Oslo, Norway) was per-formed in the lower right extremity of a healthy volunteer.The MR system and pulse sequence used were identical tothose for the phantom measurements. The volunteer wasplaced feet first in the bore. Before administration of CM, acoronal image volume was acquired using the flow-com-pensated sequence. The CM transit time from injection toimaging volume was estimated at 37 seconds using a testbolus technique and 2 ml of CM [27,28]. 38 ml of theextra-cellular CM was injected over 30 seconds followed bya normal saline flush. After that, three CE-MRA imagevolumes with coronal view and superior-inferior readoutdirection were acquired. The first acquisition was timed toonly enhance the arteries, i.e. the center of the bolus arrivedduring sampling of the central lines in k-space (half theacquisition time). This image volume was obtained usingflow-compensation. The aim of the second and third acqui-sition was to obtain image volumes with both arteries andveins enhanced, imitating a “poorly timed” first-pass acqui-sition or an acquisition with a blood pool agent. The secondacquisition was obtained using flow-compensation and thethird acquisition was obtained without flow-compensation.The delay between each acquisition was approximately 80seconds. This time was necessary for reconstruction of im-ages and storing of the raw data. Two days prior to theCE-MRA, the flow velocity in the popliteal artery of thesame volunteer was measured using the Siemens flow quan-tification package.

2.4. Volunteer measurement, blood pool agent

CE-MRA using an investigational blood pool agent (MS-325, EPIX Medical, Cambridge, MA, USA) was performedon the lower leg of a volunteer. A Philips NT 1.5 T system(software release 6.2, PowerTrak 6000 gradients, PhilipsMedical Systems, Best, The Netherlands) and the knee coilwere used. The pulse sequence used for CE-MRA imagingwas a regular spoiled 3D gradient echo (T1-FFE) sequence(TE/TR � 2.2/6.6 msec, FA � 25°, matrix � 256 � 512 �32, FOV � 280 � 400 � 38 mm3).

A CM dose of 0.07 mmol/kg was diluted to 31 ml withnormal saline and injected intravenously over 30 seconds,followed by a 30 ml normal saline flush. No pre-contrastimage volumes were acquired. Two contrast-enhanced im-age volumes were acquired during CM concentration equi-librium, approximately 45 minutes after injection. One wasacquired using flow-compensation in the readout and slicedirection and the other without flow compensation. Both

volumes were acquired with four averages for a total scanduration of 120 s.

2.5. Post-processing

All raw data was transferred to an Intel based worksta-tion running Microsoft Windows NT 4.0 and IDL software(Research Systems, Boulder, CO, USA). Both magnitudeand phase image volumes were reconstructed.

To remove all phase effects other than those caused byflow, the flow-compensated phase image volumes from thethree different experiments were subtracted from their cor-responding flow-sensitive phase image volumes. Ideallythese subtracted phase images will only contain signal inpixels with superior-inferior flow and have either positive ornegative value depending on the direction of the arterial/venous flow. The flow sensitivity of the modified pulsesequence (used in phantom measurements and volunteerexperiments with extra-cellular CM) was determined fromthe flow phantom measurements and the calculations [seeEq. (1)]. Line profiles were extracted from the image slicecentered over the two flow pipes in each subtracted phaseimage volume. The profiles were averaged over nine pixelsand centered on line 256 (of 512). The phase values in thepositions corresponding to the center of each pipe wereextracted from the respective line profiles. Since phasealiasing was present for the two highest flow velocities,these phase values were “un-wrapped”. The actual flowvelocity in the center of the pipe depends on the type of flowpresent, i.e. laminar versus turbulent. For laminar flow, thevelocity in the center of the pipe is twice the mean velocity,and in the turbulent case it is 5/4 times the mean velocity[29].

Calculations of the Reynolds number [29] indicated thatthe flow was laminar for the two lowest measured meanflow velocities (9 and 15 cm/s) and turbulent for the twohigher velocities (65 and 80 cm/s). The extracted phasevalues were plotted versus flow velocity taking the flowtypes into account.

The segmentation process for achieving venous suppres-sion is described below, and was performed the same way inboth the phantom measurements and the two types of vol-unteer measurements. Two different segmentation methodswere evaluated in the subtracted phase image volumes, asimple global segmentation, and a form of local (also calleddynamic) thresholding [30].

Global segmentation is based on a single threshold overthe entire subtracted phase image volume to determinewhich pixels should be considered an artery versus a vein.This threshold was set to zero, and all pixels with a valuelarger than zero were considered arteries and all values lessthan zero were considered veins.

In the local segmentation process, each pixel value iscompared to the mean value of its neighboring pixels (in-stead of to a global threshold value). All pixels with apositive difference were regarded arteries and all pixels with

51Jonas Svensson et al. / Magnetic Resonance Imaging 20 (2002) 49–57

a negative difference were regarded veins. Local segmen-tation was performed for three different sizes of the “neigh-borhood”: 3 � 3, 5 � 5 and 9 � 9 pixels. Each pixel withina vessel should preferably not be compared to a neighbor-hood of pixels within the same vessel type, as that coulddecrease the efficiency of the segmentation. In order toimprove the outcome of this segmentation, the neighbor-hood of pixels was chosen in the transverse plane, i.e. theplane perpendicular to the vessel direction (Fig. 1). Localsegmentation was not performed in the phantom measure-ments.

The result of either segmentation methods is that everypixel is determined to represent either an artery or a vein.Thus, in these “masks” there is no information regardingwhich of the pixels are within a vessel. To acquire thisinformation, a global threshold was applied to the contrast-enhanced magnitude image volume. The threshold was setto a value smaller than the signal in the contrast-enhancedvessels, but larger than the signal in the surrounding tissue.The value was chosen manually from the histogram of thepost-contrast magnitude volume. To extinguish the veins inthe contrast-enhanced magnitude image volume, all pixelsthat were considered both a vein and belonged within avessel were set to zero. The image volume acquired with thenon-specific extra-cellular CM was also background-sub-tracted (using the pre-contrast image volume) to suppresssignal outside the vessels.

Maximum intensity projections (MIP) of all segmentedvolumes were calculated using a built-in IDL MIP algo-rithm.

3. Results

3.1. Phase accumulation and flow

The measured phantom phase accumulation plotted ver-sus flow velocity agrees well with the calculated values(Fig. 2). The calculated velocity encoding value (VENC) ofthe pulse sequence, i.e. the flow velocity that yields a 180°phase accumulation, was 74 cm/s. The mean flow velocitypresent in the popliteal artery of the volunteer was measuredas approximately 6 cm/s (over a whole cardiac cycle), withpeaks up to 50 cm/s during systole. Theoretically this couldyield problems due to phase aliasing in the center of thevessel if the flow is laminar at peak velocity. However,since the acquisition occurs over several cardiac cycles, it isprobable that the actual phase value will reflect the meanrather than peak flow velocity. A flow velocity of 6 cm/s inthe vessel will correspond to a phase accumulation of ap-proximately 15°.

Accordingly, the chance of phase aliasing in the vesselsof interest is low.

Fig. 1. During local segmentation each pixel value was compared to the mean value of its neighbouring pixels in the transverse plane (not in the originalcoronal plane), i.e. the plane perpendicular to the flow-sensitive direction.


3.2. Segmentation

3.2.1. Flow phantomThe MIP images from the opposing flow phantom mea-

surements had similar high signal in each pipe (Fig. 3a).After arterial segmentation, the pipe with inferior flow is

unchanged whereas the pipe with superior flow is nearlyentirely suppressed (Fig. 3b). Global segmentation wasused. The result of the segmentation shown here was per-

formed on the image volume obtained when using a meanflow velocity of 9 cm/s. Segmentations were also performedon the image volumes obtained when using flow velocitiesof 15 and 25 cm/s with equally good suppression results.

3.2.2. Extra-cellular contrast mediumThe first image volume acquired after CM injection re-

sulted in good arterial enhancement with almost no visual-ization of venous structures (Fig. 4a). The volumes (flow-compensated and not flow-compensated) acquiredapproximately three minutes later (designed to simulate apoorly timed acquisition), contained arterial and venousenhancement (Fig. 4b). The overall vessel contrast in thisacquisition is somewhat lower compared with the arterialphase, due to CM distribution throughout the intravascularspace as well as into the extra-cellular space. Both theflow-compensated and the not flow-compensated imageswere background subtracted using the pre-contrast imagevolume.

Global segmentation was performed on the second imagevolume with the aim of venous suppression (Fig. 4c). Ide-ally, the result of this segmentation should make only ar-teries visible and contain no venous enhancement (similar tothe first-pass image in Fig. 4a). In the segmented image, themajor arteries are still visible and good venous suppressionis achieved. However, many of the smaller arteries are nolonger visualized, particularly when oriented in the trans-

Fig. 2. Phase values from measured flow phantom data with flow inopposite directions, positive (■ ) and negative (F). The solid lines arecorresponding calculated values. The calculated and the measured dataagree well.

Fig. 3. MIP image from flow phantom measurement with a mean flow velocity of 9 cm/s. The flow in the two pipes is oppositely directed and the signallevels are similar (a). The arrows indicate flow direction. Corresponding MIP after a global segmentation where the inferior “arterial” flow direction isunchanged and the superior “venous” flow direction is entirely suppressed (b).


verse plane. In the lower part of the image, the larger veinsare not completely suppressed.

Local segmentation was also performed on the sameimage volume using three different neighborhood sizes: 3 �3 (Fig. 5a), 5 � 5 (Fig. 5b) and 9 � 9 (Fig. 5c). Generally,the local segmentation does not seem very sensitive to thechoice of neighborhood size. However, when a 9 � 9neighborhood is used, vessel walls of the larger arteries donot appear as “pixelated” as for the other two neighborhoodsizes. A visual comparison between global segmentationand the 9 � 9 local segmentation gives the impression thatthe two methods are similarly successful.

3.2.3. Blood pool agentTwo image volumes (with and without flow-compensa-

tion) were acquired during the steady-state concentration of

CM. Arterial and venous enhancement is therefore presentin both images, and the two sets of magnitude images arenearly identical. Only the volume acquired in flow-compen-sated mode is presented here (Fig. 6a). No backgroundsuppression was applied in this image, as no pre-contrastimages were acquired.

Enhancement is present in both arteries and veins. How-ever, the patient has a moderate stenosis of the anterior tibialartery, and a high grade stenosis of the tibio-peroneal trunkwith post stenotic dilatation. This explains the lack of com-plete arterial filling and apparent irregularities of the arterialstructures. The lack of complete venous filling is likely aresult of under-distension of the veins, probably com-pounded by compression of the calf in the coil.

Segmentation was performed in attempt to suppress the

Fig. 4. MIP images from acquisitions during extra-cellular contrast medium enhancement. The first acquisition was timed to enhance only the arteries (a),whereas the aim of the acquisition in (b) was to simulate a poorly timed acquisition. Global segmentation was applied to the image volume in (b) in attemptto suppress the veins (c). Ideally, the segmented image should look like the first-pass image (a).

Fig. 5. MIP images after local segmentation trying to suppress the veins. Three different transverse neighborhood sizes were used, 3 � 3 (a), 5 � 5 (b) and9 � 9 pixels (c).


veins using both local segmentation (neighborhood size 5 �5, Fig. 6b) and global segmentation (Fig. 6c).

The incomplete vessel filling makes evaluation of theresults difficult. Nonetheless, with local segmentation themajor vessels seem correctly segmented, whereas withglobal segmentation all venous signal is correctly sup-pressed, but much of the arterial signal is also eliminated.

4. Discussion

The ideal way to acquire contrast-enhanced MR angio-grams of the lower extremity arteries would be to performMRA during equilibrium phase of a blood pool agent anduse a post-processing method able to suppress the venousenhancement. Then the need for extremely short acquisitiontimes, complicated timing procedures and bolus chasemethods would be eliminated, and the acquisition processwould be considerably simplified. Unfortunately, no suchperfect post-processing technique exists today. In this study,a method based on the differences in flow induced phaseaccumulation from opposite directional arterial (inferior)and venous (superior) flow was developed. It was evaluatedusing both a flow phantom and extra-cellular contrast me-dium with volunteers as well as in one pilot experiment witha blood pool agent. Two different MRI systems were usedin the experiments. The overall impression of the suggestedmethod was that it worked well in larger superior-inferiordirected vessels of the lower extremities regardless of thetype of contrast medium and MRI system used.

One advantage of the proposed method is that it can beused on datasets acquired with standard CE-MRA pulsesequences. In this work, the experiments performed usingthe blood pool agent were performed without changing thestandard sequences in any way. In the phantom and extra-

cellular CM experiments, the pulse sequence was onlyslightly modified to enable flow-compensation, as the orig-inal sequence did not exist in a flow-compensated version.

Another advantage to the proposed technique is thenearly automatic process of segmentation. At this time theonly user input required is to set the global threshold forextracting the vessels (both arteries and veins) from thesurrounding tissue. This step could be automated as well ifa calculation of the threshold from the magnitude imagevolume histogram or a vessel tracking/region growing al-gorithm [17,19] is used. The suggested method is based onthe requirement that all vessels are oriented parallel to theleg. Some of the smaller vessels in the leg do not fulfill thisrequirement and therefore some of the smaller arteries areincorrectly handled and appear suppressed after segmenta-tion. Furthermore, longitudinal small vessels and largerveins in the lower part of the image volume are hard tosuccessfully segment due to their low flow velocities. Thiseffect will probably be even more pronounced in patientssuffering from poor circulation (lower flow velocities). Away to compensate for this would be to increase the sensi-tivity of the flow sensitive sequence. Either using strongergradients and/or increasing TE could achieve this. Cautionmust be taken, however, not to introduce phase aliasing ifsuch modifications are employed.

In principle, the local segmentation could be performeddirectly on the phase image volume acquired with the flowsensitive sequence without using phase subtraction. How-ever, in this case phase effects from sources other thanlongitudinally directed flow may exist, for example suscep-tibility effects originating from the blood oxygenation levelor contrast medium [16,31]. The difference between arterialand venous blood oxygenation could either degrade or evenenhance the result of the flow-induced phase based segmen-tation. With the pulse sequence used in the flow phantom

Fig. 6. MIP images from acquisition post contrast enhancement with a blood pool agent. Images were acquired during steady-state of the CM (a). Two typesof segmentation were applied, local (b) and global (c). The local segmentation method was more efficient than the global.


and the extra-cellular CM experiments (TE � 3.7 msec anda static magnetic field of 1.5 T), the phase generated fromdeoxygenation of blood will be approximately 7 degrees[16], assuming an oxygenation level of 0.75 (venous blood[32]) and a hematocrit of 0.42 (in a blood vessel parallel tothe static magnetic field). Compare this with the expectedflow induced phase accumulation of 15 degrees for a flowvelocity of 6 cm/s using the same pulse sequence. Thus tominimize the risk of image degradation, it is therefore pref-erable to perform a phase subtraction.

As an alternative, instead of removing these susceptibil-ity induced phase differences using a phase subtraction, ithas been proposed to utilize them for suppression of venoussignal in the lower extremities [16]. The accumulation offlow-induced phase is either positive or negative dependingon the flow direction and the design of the gradients used.The susceptibility-induced phase accumulation is dependenton TE and differences in oxygentation level. It should intheory be possible to sum these two different phase effectsby designing the flow sensitive gradients with this in mind.The result would be an increased arterial and venous phasedifference, and a more stable technique than for each tech-nique alone.

While the presence of a contrast medium also introducessusceptibility changes, at the short echo times used in thisexperiment (TE � 3.7 msec), these effects should be rathersmall [16].

The global segmentation was not as successful as thelocal segmentation when applied to the image volume ac-quired using the blood pool agent. A global segmentation istotally dependent on a successful phase subtraction. If thephase subtraction fails in any way (e.g. due to patientmotion between acquisitions), the global threshold that isused may no longer be valid and the segmentation fails. Thismay be one reason for the poor blood pool agent result usingglobal segmentation. Accordingly, if a successful phasesubtraction cannot be assured, local segmentation is thepreferred method.

Some other segmentation methods suggested in the lit-erature rely on the differences in the temporal enhancementpatterns of arteries and veins following a CM bolus injec-tion [20–23]. All of these methods require imaging duringthe first pass of the bolus, and cannot be used if images areacquired only during steady state concentration of a bloodpool agent. In a situation where the arterial structures of theentire leg will be investigated, several acquisitions areneeded together with movement of the patient support table.Accordingly, it may be impractical to follow the dynamicenhancement pattern of both arteries and veins in eachacquisition position, making these methods unsuitable.

If imaging is performed during steady state concentrationof a blood pool agent, the total time available for imaging islong. This longer imaging time can be used to acquireimages with an increased spatial resolution [12]. If thespatial resolution is adequate, arteries and veins can poten-tially be separated through vessel tracking algorithms [17–

19]. Assuming these methods are feasible, they still need astarting point for the region growing. The flow-based seg-mentation used in this study may be useful to automaticallygenerate such starting points.

While the single blood-pool image demonstrates the fea-sibility of this approach for this type of contrast agent, anyoptimization for clinical exams has yet to be performed.Many questions remain: global vs. local segmentation, sizeof optimal local neighborhood, the need for potential addi-tional flow encoding or small increases in echo time overthose present in standard sequences, as well as the ability toadditionally use phase shifts due to oxygenation differencesbetween arterial and venous blood. However, even in thepresence of these questions the primary conclusion of thiswork remains: using conventional 3D-MRA sequences,there may be sufficient velocity phase contrast to allowsegmentation of veins from arteries in contrast-enhanced legMR angiograms.

5. Conclusion

A new method for separating arteries from veins in thelower extremities was developed based on flow-inducedphase effects. This technique was investigated using imagesobtained with 3-D CE-MRA pulse sequences. The sug-gested method proved capable of suppressing most venoussignal while leaving the larger arteries unchanged. In ves-sels not oriented longitudinal to the legs and small vesselswith low flow velocity, however, the segmentation was lesssuccessful. The current results could potentially be im-proved by either making the sequence more flow sensitive,or combining this method with other techniques such asthose based on susceptibility differences and/or vesseltracking algorithms.

Acknowledgment

The authors wish to express their thanks to Dr CarstenThomsen, Department of Radiology, Rigshospitalet, Copen-hagen, Denmark, for valuable discussions about segmenta-tion techniques in MR angiography.

References

[1] Czum JM, Ho VB, Hood MN, Foo TK, Choyke PL. Bolus-chaseperipheral 3D MRA using a dual-rate contrast media injection. JMagn Reson Imaging 2000;12:769–75.

[2] Grist TM. MRA of the abdominal aorta and lower extremities. JMagn Reson Imaging 2000;11:32–43.

[3] Ho KY, Leiner T, de Haan MW, Kessels AG, Kitslaar PJ, vanEngelshoven JM. Peripheral vascular tree stenoses: evaluation withmoving-bed infusion-tracking MR angiography. Radiology1998;206:683–92.


[4] Lee HM, Wang Y. Dynamic k-space filling for bolus chase 3D MRdigital subtraction angiography. Magn Reson Med 1998;40:99–104.

[5] Meaney JF, Ridgway JP, Chakraverty S, Robertson I, Kessel D,Radjenovic A, Kouwenhoven M, Kassner A, Smith MA. Stepping-table gadolinium-enhanced digital subtraction MR angiography of theaorta and lower extremity arteries: preliminary experience. Radiology1999;211:59–67.

[6] Anzai Y, Prince MR, Chenevert TL, Maki JH, Londy F, London M,McLachlan SJ. MR angiography with an ultrasmall superparamag-netic iron oxide blood pool agent. J Magn Reson Imaging 1997;7:209–14.

[7] Engelbrecht MR, Saeed M, Wendland MF, Canet E, Oksendal AN,Higgins CB. Contrast-enhanced 3D-TOF MRA of peripheral vessels:intravascular versus extracellular MR contrast media. J Magn ResonImaging 1998;8:616–21.

[8] Kroft LJ, de Roos A. Blood pool contrast agents for cardiovascularMR imaging. J Magn Reson Imaging 1999;10:395–403.

[9] Lauffer RB, Parmelee DJ, Dunham SU, Ouellet HS, Dolan RP, WitteS, McMurry TJ, Walovitch RC. MS-325: albumin-targeted contrastagent for MR angiography. Radiology 1998;207:529–38.

[10] McLachlan SJ, Morris MR, Lucas MA, Fisco RA, Eakins MN,Fowler DR, Scheetz RB, Olukotun AY. Phase I clinical evaluation ofa new iron oxide MR contrast agent. J Magn Reson Imaging 1994;4:301–7.

[11] Saeed M, Wendland MF, Engelbrecht M, Sakuma H, Higgins CB.Value of blood pool contrast agents in magnetic resonance angiog-raphy of the pelvis and lower extremities. Eur Radiol 1998;8:1047–53.

[12] Grist TM, Korosec FR, Peters DC, Witte S, Walovitch RC, Dolan RP,Bridson WE, Yucel EK, Mistretta CA. Steady-state and dynamic MRangiography with MS-325: initial experience in humans. Radiology1998;207:539–44.

[13] Hany TF, McKinnon GC, Leung DA, Pfammatter T, Debatin JF.Optimization of contrast timing for breath-hold three-dimensionalMR angiography. J Magn Reson Imaging 1997;7:551–6.

[14] Ho VB, Choyke PL, Foo TK, Hood MN, Miller DL, Czum JM, AisenAM. Automated bolus chase peripheral MR angiography: initial prac-tical experiences and future directions of this work-in-progress. JMagn Reson Imaging 1999;10:376–88.

[15] Strouse PJ, Prince MR, Chenevert TL. Effect of the rate of gado-pentetate dimeglumine administration on abdominal vascular andsoft-tissue MR imaging enhancement patterns. Radiology 1996;201:809–16.

[16] Wang Y, Yu Y, Li D, Bae KT, Brown JJ, Lin W, Haacke EM. Arteryand vein separation using susceptibility-dependent phase in contrast-enhanced MRA. J Magn Reson Imaging 2000;12:661–70.

[17] Cline HE, Dumoulin CL, Lorensen WE, Souza SP, Adams WJ.Volume rendering and connectivity algorithms for MR angiography.Magn Reson Med 1991;18:384–94.

[18] Cline HE, Thedens DR, Meyer CH, Nishimura DG, Foo TK, LudkeS. Combined connectivity and a gray-level morphological filter inmagnetic resonance coronary angiography. Magn Reson Med 2000;43:892–5.

[19] Lin W, Haacke EM, Smith AS, Clampitt ME. Gadolinium-enhancedhigh-resolution MR angiography with adaptive vessel tracking: pre-liminary results in the intracranial circulation. J Magn Reson Imaging1992;2:277–84.

[20] Bock M, Schoenberg SO, Floemer F, Schad LR. Separation of arter-ies and veins in 3D MR angiography using correlation analysis. MagnReson Med 2000;43:481–7.

[21] Du J, Mazaheri Y, Carroll TJ, Esparsa Coss E, Grist TM, MistrettaCA. Autmated region-specific VTRAC segmentation on peripheralMRA. Proc ISMRM, 8th Scientific Meeting. Denver: 2000. p. 1807.

[22] Kaandorp DW, Kopinga K, Wijn PF. Venous signal suppression in3D dynamic Gd-enhanced carotid artery imaging using the eigenim-age filter. Magn Reson Med 1999;42:307–13.

[23] Mazaheri Y, Carroll TJ, Korosec FR, Block W, Du J, Grist TM,Mistretta CA. High resolution CE-MRA using dual-resolution acqui-sition and segmentation based on spatial frequency-dependent 2Dtemporal correlation. Proc ISMRM, 8th Scientific Meeting. Denver:2000. p.189.

[24] Svensson J, Leander P, Ståhlberg F, Thomsen C, Sjoberg S, OlssonL-E. Segmentation of arteries from veins in contrast-enhanced 3-Dmagnetic resonance angiography of the lower extremities. ProcISMRM, 8th Scientific Meeting. Denver: 2000. p. 1809.

[25] Svensson J, Leander P, Ståhlberg F, Thomsen C, Sjoberg S, OlssonL-E. Can the inherent phase information in the MR image be used toseparate arteries from veins in CE-MRA of the lower extremities?Proc ESMRMB 16th Annual Meeting. Seville: 1999. p. 100.

[26] Moran PR. A flow velocity zeugmatographic interlace for NMRimaging in humans. Magn Reson Imaging 1982;1:197–203.

[27] Earls JP, Rofsky NM, DeCorato DR, Krinsky GA, Weinreb JC.Breath-hold single-dose gadolinium-enhanced three-dimensional MRaortography: usefulness of a timing examination and MR powerinjector. Radiology 1996;201:705–10.

[28] Steffens JC, Link J, Grassner J, Mueller Huelsbeck S, Brinkmann G,Reuter M, Heller M. Contrast-enhanced, K-space-centered, breath-hold MR angiography of the renal arteries and the abdominal aorta. JMagn Reson Imaging 1997;7:617–22.

[29] Lefebvre XP, Pedersen EM, Hjortdal JO, Yoganathan AP. Hemody-namics. In: Lanzer P, Yoganthan AP, editors. Vascular imaging bycolor doppler and magnetic resonance. Berlin: Springer-Verlag, 1990:p. 51–7.

[30] Rogowska J. Overview and fundamentals of medical image segmen-tation. In: Bankman IN, editors. Handbook of medical imaging.Processing and analysis. London: Academic Press, 2000: p. 69–85.

[31] Hoogenraad FG, Reichenbach JR, Haacke EM, Lai S, Kuppusamy K,Sprenger M. In vivo measurement of changes in venous blood-oxygenation with high resolution functional MRI at 0.95 tesla bymeasuring changes in susceptibility and velocity. Magn Reson Med1998;39:97–107.

[32] Albert MS, Huang W, Lee JH, Patlak CS, Springer CS. Susceptibilitychanges following bolus injections. Magn Reson Med 1993;29:700–8.


Separation of arteries and veins using flow-induced phase effects in contrast-enhanced MRA of the...

Documents

Transcript of Separation of arteries and veins using flow-induced phase effects in contrast-enhanced MRA of the...