173

Detection of Thrombus by Using

Phase-Image MR Scans: ROC Curve
Analysis

L;: ‘:... ‘ ,

Nuno J. Tavares1 Spin-echo phase images have been shown to be sensitive to blood flow and have
Wolfgang Auffermann been used to differentiate slow flow from thrombus, with an apparent advantage in
Jeffrey J. Brown comparison with spin-echo intensity images alone. In order to quantify the diagnostic
Thomas J. Gilbert efficacy of phase images, a study was performed comparing the sensitivity and speci-
ficity of MR imaging in identifying intravascular thrombus using spin-echo magnitude
Christian Sommerhoff
images alone and combined wIth phase images. In 45 subjects, 66 vessels with
Charles B. Higgins questionable Intraluminal signal were reviewed in a blinded manner by four radiologists
using seven levels of certitude for the diagnosis of thrombus. Vessels in the thorax,
abdomen, and pelvis were included in the evaluation and were selected on the basis of
the presence of intraluminal signal, which raised the possibility of intravascular disease.
Corroborative studies were available in all cases. Receiver-operator-characteristic
curves were constructed for the accuracy of the decision of intraluminal thrombus vs
flow signal when using magnitude images alone and when using magnitude plus phase
images. Magnitude images identified thrombus with a sensitivity of 35% at a specificity
of 90%. On the other hand, combining magnitude image and phase images yielded
sensitivities of 85% and 83% at specificities of 90% and 95%, respectively.
We conclude that addition of phase images substantially increases the level of
confidence in detecting intravascular thrombosis.

AJR 153:173-178, July 1989

The variable appearance of intraluminal signal due to various flow phenomena

may render spin-echo MR images equivocal in the distinction between flow signal
and intraluminal thrombus. Previous reports have described the use of phase
imaging to distinguish thrombus from slow flow [1 -3]. Although phase images have
been shown to be effective in making this distinction in a small series of patients,
no studies have provided a systematic quantitative evaluation of the diagnostic
accuracy of this technique in comparison with the standard spin-echo intensity
image.
A study was performed to determine the sensitivity and specificity of MR phase
images in showing the presence
or absence of occluded (thrombus-filled) vessels.
A moderate number of vessels (n = 66) were evaluated for the presence of
thrombus by using magnitude and phase images, and receiver-operating-charac-
Received January 5, 1989; accepted after revi- tenstic (ROC) curve analysis was done to assess the added diagnostic efficacy of
sion February 24, 1989. the phase image. By using different levels of confidence, sensitivity and specificity
‘All authors: Department of Radiology (C309, for the appearance of thrombus were determined for magnitude images alone and
Box 0628), lkiiversity of California, San Francisco
Medical School, San Francisco, CA 94143. Address then for magnitude and phase images together.
reprint requests to C. B. Higgins.
N. J. Tavares is on leave from the Dept. of Materials and Methods
Radiology, Hospital Santa Marta, H. C. L., Lisbon,
Portugal, and is supported by individual grant Subjects
#Proc. 3.3/P. 517 from the Luso-American Devel-
opment Foundation, Lisbon, Portugal. Forty-five subjects (40 patients and five normal volunteers) underwent MR imaging to
0361-803x/89/1531-0173 determine vascular patency in one or more vessels. The 29 males and 1 6 females ranged in
American
#{174} Roentgen Ray Society age from 5 to 80 years. A total of 66 vessels were evaluated: 40 arteries, 23 veins, and three
174 TAVARES ET AL. AJR:153, July 1989

heart studies were included. Eight vessels (seven arteries, one vein) dimensional Fourier transform imaging method was used to recon-
were evaluated in the chest region, 33 in the abdominal region (11 struct images.
arteries, 22 veins), and 22 in the pelvic region (22 arteries). Phase imaging-The basic principles of phase imaging have been
In 1 3 patients, thrombus was shown by one or more corroborating described [4]. In brief, the measured quantity is the magnetization of
studies (surgery in five patients, angiography in five, CT in two, the hydrogen nuclei within the tissue. Whereas conventional MR
sonography in one) (Table 1). images reflect the amplitude of the magnetization, the phase recon-
The presence of thrombus was ruled out in 27 patients by means struction reflects the direction.
of at least one corroborating study (autopsy, one patient; surgery, By design, MR sequences result in the same phase angle through-
seven; angiography, 12; CT, six; sonography, one). These patients out the image unless interrupted by flow, field inhomogeneity, or
had many preevaluation diagnoses, including aortobifemoral bypass susceptibility differences. Typical gradient waveforms used to image
graft occlusion in seven patients, aortic aneurysms in five, aortic stationary tissue produce a phase shift proportional to velocity in the
dissection in two, renal vein thrombosis in three, left or right ventric- gradient direction for constant gradient amplitude and duration [5].
ular thrombus in three, IVC thrombus in two, aortoiliac occlusive Phase is a periodic quantity and therefore is not displayed accu-
disease in two, pulmonary embolism in one, and pulmonary arterio- rately by typical linear gray scales. For that reason, the equivalent
venous malformation in one. In addition, five normal volunteers (five angles -1 80#{176}
and +1 80#{176}
are represented by black and white, re-
men, 33-38 years old) were interposed in the study. suIting in a sharp transition in the image.
In practice, stationary nuclei are
assigned to the middle shade of
gray, and moving spins appear increasingly white or black. Because
of the cyclic nature of the gray scale, extremes of black and white
Imaging Technique both indicate 1 80#{176} phase shift. The technique is sensitive to either
antegrade or retrograde flow in a vessel. Apart from motan, the local
MR imaging was performed by using a 0.35-T superconducting field experienced by a nucleus may vary as a result of intrinsic
magnet system (Diasonics, Milpitas, CA). inhomogeneities in the polarizing field (field inhomogeneity) and the
Spin-echo images-Images were acquired by using spin-echo tissue magnetization that is induced by the polarizing field (magnetic
(SE)imaging technique. Ti-weighted images were generated by using susceptibility). The presence of these factors can result in phase
SE pulse sequences with 500/30 (TRITE), and T2-weighted images shifts that may be erroneously interpreted as flow. Regions with air
were acquired at 2000/30-60. (intestine, lungs) can give inaccurate impressions of phase shift.
A multislice imaging technique was used. Multisection single- and
dual-echo images in the transverse plane were obtained in all studies.
Additional coronal planes were obtained in nine studies and sagittal
planes in three studies. Imaging Analysis
ECG gating was used in 13 patients (TE = 30, TR = RR interval
ofthe patient’s electrocardiogram). A multislice ECG-gated technique MR images were reviewed retrospectively by four radiologists
was used so that images at any level were in either systole or diastole blinded to the clinical diagnosis and the results of corroborative
but were not acquired at both planes of the cardiac cycle. Section studies. Images were selected by one of the authors, who did not
thickness was 1 0 mm, with no gap or a 1 -mm gap. The two- participate in the reading.

TABLE 1: Patients with Thrombus

MR Analysis Corroborative Studies

Patient Age
No. (yr) ex Diagnosis Technique
(certitude)
1 24 F Thrombus (1 00%) Thrombus (1 00%) Portal vein thrombosis CT
2 76 M Thrombus possibly Flow (0#{176}h
thrombus) Aortobifemoral graft/ Angiography
present (60%) right limb occlusion
3 75 F Thrombus probably Flow (0% thrombus) Aortobifemoral graft/left Angiography
absent (20%) iliofemoral occlusion
4 26 M Uncertain (50%) No flow (1 00%) Chronic lung hematoma Surgery
5 54 M Thrombus (1 00%) Thrombus (1 00%) IVC thrombus Surgery
6 18 M Thrombus possibly Thrombus (1 00%) Renal vein thrombus Sonography
present (60%)
7 28 M Thrombus probably Thrombus (1 00%) Aortic coarctation/ Surgery
present (80%) thrombus
8 60 M Thrombus possibly Thrombus probably IVC thrombus CT
absent (40%) present (80%)
9 53 M Thrombus (1 00%) Thrombus (1 00%) Renal vein and IVC Surgery
thrombus
10 33 M Thrombus (1 00%) Thrombus (1 00%) Right portal vein throm- Angiography
bus
11 69 M Thrombus possibly Thrombus (1 00%) Aortic graft infection/ Angiography
present (60%) occlusion
12 79 F Thrombus possibly Thrombus (1 00%) Aortic dissection/false- Surgery
present (60%) lumen thrombus
13 5 F Uncertain (50%) No flow (1 00%) Right pulmonary artery Angiography
Note.-IVC = in ferior vena cava.
AJR:153, July 1989 MR DETECTION OF THROMBUS 175

In the first step, the magnitude images were assessed and a dence level three, 50% certainty); in one patient thrombus
confidence level for the presence of intraluminal pathological signal was possibly absent (confidence level two, 60% normal); and
compatible with thrombus was assigned. The criteria used for the finally, in one patient thrombus was probably absent (confi-
presence of intraluminal thrombus on magnitude images were (i ) the
dence level one, 80% normal) (Fig. 3).
absence of a dark rim around the central zone of intraluminal signal;
Conversely, after review of magnitude and phase images
(2) the presence of intraluminal signal in both first and second echo
together, thrombus was shown with high certainty in 1 1 of
images; and (3) when several planes were available, the presence of
intraluminal signal in both planes. 1 3 patients (Fig. 3). In 1 0 patients, the diagnosis was definite
In the second step, phase and magnitude images together were (confidence level six, 1 00% certainty), and in one patient the
analyzed by the same panel of radiologists 2 days to 1 week later. diagnosis was probable (confidence level five, 80% certainty).
The criterion to show thrombus on phase images was the absence An example of magnitude and phase images showing throm-
of phase shift of the magnetization vector in the lumen of the vessel bus with a high confidence level is shown in Figure 1 . How-
of interest. On the phase image, intraluminal thrombus will have the ever, in two patients with thrombus confirmed by corrobora-
same intensity pattern as the surrounding stationary tissue, whereas tive studies, the use of phase images resulted in a false-
flow will be evident by a different gray level compared with the negative diagnosis (patients 2 and 3, Table 1). Both studies
stationary tissue. Finally, a confidence level for the presence of
were performed in order to assess aortofemoral bypass graft
intraluminal thrombus combining phase and magnitude analysis was
assigned for each case, for the following observations: (a) presence
patency. Both exams were interpreted as normal flow where
or absence of thrombus on magnitude images and (b) presence or there was an occluded vessel, as determined by angiography.
absence of thrombus (occlusion) on magnitude images plus phase The absence of thrombus was shown in 53 vessels in 32
images. patients by the corroborative studies. On the basis of review
For each case, the true diagnosis was established on the basis of of the magnitude images alone (Fig. 3B), only 27 vessels were
the findings of autopsy, surgery, angiography, CT, or sonography correctly determined as definitely patent (confidence level
(echocardiography), with preference for certitude in that order. zero, 100% certainty), six as probably patent (confidence
level one, 80% certainty), and four as possibly patent (confi-
dence level two, 60% certainty). A diagnosis could not be
Statistical Analysis
made for six vessels. A false-positive diagnosis for the pres-
In consensus, the panel of reviewers assigned a confidence level ence of thrombus was made in 10 vessels.
for the diagnosis. The responses were categorized in six different Conversely, when using both phase and magnitude images,
levels: level six, thrombus definitely present (reviewers were 100% 48 of 53 vessels were diagnosed correctly as patent (level
certain); level five, thrombus probably present (80% certain); level zero and one) (Fig. 3B). Only one unsure and four false-
four, thrombus possibly present (60% certain); level three, unsure positive diagnoses of thrombus were made by using phase
(50% certain); level two, possibly absent (60% certain); level one, images. From these, only three patients had the diagnosis of
thrombus probably absent (80% certain); level zero, thrombus defi-
definite or probable thrombus (confidence levels six and five,
nitely absent (1 00% certain).
ROC curves [6, 7] were generated from these data at each
100% and 80% certainty). In the other two patients, thrombus
confidence level. For confidence level six, the true-positive fraction was possibly present in one (confidence level four, 60%
was calculated as the number of true-positive responses divided by certainty), and in the other patient the presence of thrombus
the number of actually positive cases, and the false-positive fraction was uncertain (confidence level three, 50% certainty).
was calculated as the number of false-positive cases divided by the An example of false-positive diagnosis of thrombus when
number of actually negative cases. The true-positive fraction and the using magnitude and phase images included a patient with a
false-positive fraction were plotted on the ordinate and the abscissa, huge benign liver mass compressing the IVC, where no flow
respectively, defining the first point of the ROC curve. could be detected on either magnitude or phase images (Fig.
The second point was generated by combining level-six and level- 2). Although no thrombus was in the lumen of the vessel,
five responses and again calculating true-positive and false-positive
phase images did not detect any flow, owing to total collapse
fractions. This procedure was repeated as responses at all levels
were progressively combined. Sensitivity (equal to the true-positive of the lumen. Figure 4 shows the ROC curves for the diag-
fraction) at specificities (equal to one minus the false-positive fraction) nosis based on the review of magnitude images alone (Fig.
of 90% and 95% was determined from the generated ROC curves 4A) or of magnitude and phase images (Fig. 4B). For magni-
by linear interpolation between the measured data points. tude images alone, the sensitivity was only 35% at a specific-
ity of 90%; a specificity of 95% was not reached. For magni-
tude and phase images combined, the sensitivity was 85% at
Results a specificity of 90% and 83% at a specificity of 95%.

Thrombus was diagnosed in 1 3 vessels of 13 patients by

corroborative studies (Table 1). Magnitude images raised the
Discussion
possibility of intraluminal thrombus in all 13 patients (Figs. 1
and 2). With magnitude images alone, the diagnosis was The presence of intraluminal thrombus has both therapeutic
considered definite in four patients (confidence level six, 100% and prognostic importance. To be suitable as a screening
certainty); in one patient the diagnosis was probable (confi- technique for the presence of intraluminal thrombus, an im-
dence level five, 80% certainty); in four patients the diagnosis aging technique ideally should have a high sensitivity and
was considered possible (confidence level four, 60% cer- specificity for this condition. Angiography is conventionally
tainty); in two patients the diagnosis was uncertain (confi- regarded as the standard diagnostic technique for intraluminal
176 TAVARES ET AL. AJR:153, July 1989

Fig. 1.-Coronal magnitude (SE 500/30) and

phaselmages of a patient with portalveln throm-
heels.
A, Signal filling portal vein Is Interpreted as
thrombus on magnitude image (arrows).
B, Corresponding phase image. Absence of
phase shift in portal vein causes It to be lndlstin-
guishable from surrounding liver, which mdi-
cafes thrombus. Phase shift in Inferior vena cava
(arrows) Is detectable from stationary surround-
ing tissue, which indicates blood flow. Apparent
abrupt termination of inferior vena cava Is
caused by vessel coursing dorsal to the confines
of the shea at this level.

A B

c---

Fig. 2.-A, Sagittal image (SE 2000/60). Be-

nign liver tumor (asterisk) compressing Inferior
vena cava raises possibility of thrombus of in-
ferlor vena cava (arrows).
B, Corresponding phase image shows no
phase shift, consistent with presence of throm-
bus In inferior vena cave (arrow). Phase shift is
evident more caudally in vessel. No thrombus
was found in inferior vena cava at surgery.
A B

50

40

U)
C,)

a) 30
a)
U) U)
C/) U)
a) a)
> > 20

10

0 I.
6 5 4 3 2 1 0 6 5 4 3 2 1 0
Confidence Level Confidence Level
A B

Fig. 3.-Frequency of confidence levels assigned for diagnosis of thrombus on basis of review of magnitude images alone (solid bars) or of magnitude
and phase images (hatched bars).
A, Patients with thrombus diagnosed by corroborative studies.
B, Patients with patent vessels.
For definition of confidence levels, see Materials and Methods.
AJR:153, July 1989 MR DETECTION OF THROMBUS 177

0.6’
TPF TPF
0.4

0.2

0.0
0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0
FPF FPF
A B

Fig. 4.-Receiver-operator-characteristic curves for diagnosis of thrombus by review of magnitude images alone (A) and by review of magnitude and
phase images (B). TPF = true-positive fraction; FPF = false-positive fraction.

thrombus. Because of the invasive nature of the technique, results of this study show that magnitude images identified
angiography is not ideal. intraluminal signal, raising the possibility of thrombus in all 13
Early experience has shown that vascular imaging is pos- vessels proved to have thrombus. However, with magnitude
sible with MR imaging [8]. Rapidly flowing blood generally images alone, thrombus was diagnosed at a high confidence
produces a signal void on SE MR images so that intraluminal level (levels six, five, four) in only nine patients. In the other
disease is usually discernible. Intravoxel dephasing and rapid four patients, magnitude images were either uncertain or were
section transition contribute to the signal loss generally ob- more favorable for the presence of slow flow (Fig. 1A). Inter-
served on SE images [5]. Despite the signal void observed in pretation of magnitude and phase images produced a high
most MR imaging sequences, in some cases, intraluminal confidence level (1 00-80%) for the presence of thrombus in
signal can be identified persistently on both Ti- and T2- 1 1 of the 1 3 vessels proved to have thrombus (Fig. 1 A).
weighted images, with signal characteristics similar to throm- Normal flow was present in 53 vessels. Based on magni-
bus [1-3, 9]. tude images alone, the presence of flow was identified with
Specific patterns of intraluminal signal, such as the pres- high probability (confidence levels zero, one, two) in only 37
ence of a dark ring (signal void) around a central zone of instances. The other 1 6 were either uncertain (confidence
signal on first echo replaced by high signal on second echo, level three) or false positive for the presence of thrombus. On
change of intravascular signal with orthogonal planes, and the contrary, the evaluation of the 53 normal vessels with
presence of rephasing phenomena, have been used to dis- phase images resulted in correct diagnosis in 48 (confidence
criminate between flow and thrombus [2, 5, 9]. Quantitative levels zero and one). In the other five, one was uncertain and
measurements of T2 values have been advocated. Large T2 four were interpreted as false positive for thrombus (Fig. 1 B).
or even negative T2 values seem to relate to the presence of The high level ofcertainty achieved in this study in detecting
flowing blood [2]. However, these measurements are time- vessel occlusion by using phase images, make phase-image
consuming and difficult to apply. Moreover, they are some- MR scanning suitable as an alternative method to address
what system dependent and have to be assessed tediously qualitatively the problem of vessel patency. Because analysis
for each individual MR unit. was done in a blinded manner, history of the patient was not
The use of phase imaging is indicated as an alternative used in the interpretation. Phase images were only equivocal
method for the evaluation of flow [1 -3]. The imaging of phase in four cases (false positive for thrombus) compared with 16
shift of the magnetization vector, which occurs for tissues cases when using magnitude images alone. Thus, a substan-
moving relative to the magnetization vectors of surrounding tially higher sensitivity and specificity can be expected with
stationary tissue, reflects vector direction of moving spins. the use of phase images compared with magnitude images
The phase shift is proportional to gradient strength and half alone. However, some pitfalls remain: resolution is a problem
of echo time (TE), assuming constant velocity motion [4, 10]. because very small vessels (i.e., iliac-femoral region) might be
In this study, a moderate number of vessels containing difficult to evaluate for the presence of thrombus, especially
intraluminal signal on SE images were evaluated for the in the regions where susceptibility problems predominate. In
presence of thrombus, by using either magnitude images or that case, alternative MR techniques such as gradient echo
magnitude and phase images combined. All were blindly sequences [1 1 , 12] might be more useful.
reviewed and in all, corroborating studies were available. The Another problem with using phase images arises in vessels
178 TAVARES ET AL. AJR:153, July 1989

with very slow flow, such as veins, where the venous return of these sequences in detecting vessel patency in a large
is very slow or when in certain circumstances images are number of clinical situations. In the future, the problem of
acquired during the diastolic phase of the cardiac cycle. As intraluminal thrombus may be addressed by more than one
phase shift is proportional to velocity and gradient strength, technique.
it is at least theoretically possible to have a very low range of
velocities where the phase shift is not visually detectable at
the normal window levels. This is especially true when flow REFERENCES
runs in the direction of the phase encoding gradient, resulting 1. White EM, Edelman RR, Wedeen VJ, Brady TJ. Intravascular signal in MR
in less phase shift than in the direction of the readout gradient. imaging: use of phase display for differentiation of blood-flow signal from
Finally, some areas might be extremely difficult to evaluate intraluminal disease. Radiology 1986;161 :245-249
2. von Schulthess GK, Augustiny N. Calculation of T2 values versus phase
because of movement. This is the case of the heart, where a
imaging for the distinction between flow and thrombus in MR imaging.
thrombus can be undetectable on phase images because all Radiology 1987;164:549-554
cavities are moving during the cardiac cycle, resulting in a 3. Rumancik WM, Naidich DP, Chandra R, et al. Cardiovascular disease:
motion-induced phase shift for the thrombus. evaluationwith MR phase imaging. Radiology 1988;166:63-68
So far, no single method is totally accurate for determining 4. van Dijk P. Direct cardiac NMR imaging of heart wall and blood flow
velocity. J Comput Assist Tomogr 1984;8:429-436
vessel patency. The use of single-slice or multislice gradient-
5. von Schulthess K, Higgins CB. Blood flow imaging with MR: spin-phase
echo images for defining flow is an interesting approach, but phenomena. Radiology 1985;157:687-695
this technique has not been evaluated systematically. Even 6. Metz CE. ROC methodology in radiology imaging. Invest Radiol 1986;
with the gradient-echo technique, it is theoretically possible 21 :720-733

that slow flow in a vessel also may be erroneously interpreted 7. Lusted LB. General problems in medical decision-making with comments
on ROC analysis. Semin Nuc! Med 1978;8:299-306
as thrombus depending on saturating conditions.
8. Axel L. Blood flow effects in magnetic resonance imaging. AiR 1984;
The performance of phase images in this study is consistent 143:1157-1166
with some early experience in the use of this technique 9. Valk PE, Hale JD, Crooks LE, et al. MRI of blood flow: correlation of image
[2, 3]. The current study quantitatively assesses the added appearance with spin-echo phase shift and signal intensity. AJR 1986;
146:931 -939
diagnostic accuracy of phase images, which can be a valuable
10. Wendt RE III, Murphy PH, FordJJ, Bryan RN, BurdineJA. Phase alterations
adjunct to SE magnitude images. Phase images have the ofspin echoes by motion along magnetic field gradients. Magn Reson Med
advantage of adding no extra time to the image sequence 1985;2:527-533
because, with the use of quadrature detection, phase infor- 11 . Haase A, Frahm J, Matthaei D, Hbnicke W, Merboldt K-D. FLASH imaging.
mation is retrieved from the measured signals. Limitations Rapid NMR imaging using low flip-angle impulses. J Magn Rca 1986;67:
258-266
remain, however, and alternative pulse sequences such as
12. Wehrh PM. Fast-scan imaging: principles and contrast phenomenology. In:
gradient echo sequences are desirable in certain circum- Higgins CB, Hricak H, eds. Magnetic resonance imaging of the body. New
stances. So far, no other study has determined the accuracy York: Raven, 1987