Volumetric analyses of central nervous system neoplasm based on MRI

Volumetric Analyses of Central Nervous System Neoplasm Based on MRI P a u l i n e A. F i l i p e k , M D * , D a v i d N . K e n n e d y , P h D * + , a n d V e r n e S. C a v i n e s s , Jr, M D , D P h i l *

Morphometric analyses were performed using an objective semiautomated algorithm on 5 sequential threedimensional Tl-weighted magnetic resonance imaging scans of a metastatic choroid plexus carcinoma, concurrent with a course of chemotherapy. The 5 scans were positionally normalized in a three-dimensional coordinate system for uniform definition of the borders of the mass. Volumes were calculated for the gadolinium-DTPA enhancing and nonenhancing cystic-appearing regions. Volumetric changes of up to 145% were measured using this method which were associated with changes in the calculated (spherical) radii of only up to 2.7 mm. Volumetric changes of up to 59% were not appreciated by visual inspection, most probably due to irregular borders and positional variability across the scans. Volumetric analyses were also performed on the right cerebeUar hemisphere, producing a 1.83% coefficient of variability across the 5 scans. The growth rates of this mass were estimated from the sequential computations, permitting in vivo observations on tumor behavior otherwise not obtainable. These analyses demonstrate the potential of this morphometric method to detect significant volumetric changes, and illustrate its use to define in vivo the growth properties of central nervous system tumors in response to therapeutic interventions. Filipek PA, K e n n e d y D N , C a v i n e s s V S Jr. V o l u m e t r i c analyses o f central nervous system n e o p l a s m based on M R I . Pediatr Neurol 1991;7:347-51.

Introduction M a g n e t i c r e s o n a n c e i m a g i n g ( M R I ) is currently the m o s t sensitive m e a n s for in vivo central nervous system ( C N S ) t u m o r detection and localization [1-4]. Particularly w h e n a u g m e n t e d by the use o f p a r a m a g n e t i c contrast agents, such as g a d o l i n i u m - D T P A , M R I has b e e n d e m o n strated to define t u m o r margins with greater clarity than other i m a g i n g modalities [ 1-4] and m a y reveal features of internal t u m o r architecture (i.e., vessels, necrosis, cysts) [3,5], potentially aiding t u m o r classification [6]. A l t h o u g h

From the *Pediatric Neurology and tRadiology Services; Center for Morphometric Analysis, Neuroscience Center; Massachusetts General Hospital; Harvard Medical School; Boston, Massachusetts. Presented in part at the Eighteenth Annual Meeting of the Child Neurology Society, San Antonio, Texas, October, 1989.

based on manual outlining techniques, prior reports h a v e demonstrated that quantitative assessments o f n e u r o i m a g ing studies [7-10] are a sensitive indicator of t u m o r volumetric change. Sequential quantitative analyses o f M R / scans can, in principle, provide additional in vivo estimates o f growth rates [ 11-12], which are accepted as the parameter that most reliably correlates with malignant potential, and best characterizes the response to treatment [ 13-19]. We report the application o f an objective s e m i a u t o m a t e d m e t h o d o f m o r p h o m e t r i c analysis to sequential gadolini u m - e n h a n c e d M R I scans. T h e s e scans were p e r f o r m e d on a 3-year-old child during concurrent c h e m o t h e r a p y for a highly m a l i g n a n t choroid plexus carcinoma. The characterization o f the b e h a v i o r o f this mass illustrates the sensitivity o f m o r p h o m e t r i c analysis and the potential for its use to estimate in vivo the g r o w t h properties o f C N S tumors or response to therapeutic interventions.

Methods Clinical History. The patient had a choroid plexus carcinoma diagnosed at 20 months of age. The tumor, which extended through the lateral ventricles of both hemispheres, was surgically excised. Following 4 postsurgical pulses of cis-platinum, no residual mass was evident on imaging studies. Two additional pulses of the infant Pediatric Oncology Group (POG) protocol (i.e., cyclophosphamide/vincristine and etoposide/cis-platinum) were completed by age 32 months. A right central facial palsy and right hemiparesis appeared at age 36 months. Computed tomographic (CT) and MRI studies confirmed recurrence of tumor in the left juxtaventricular region, extending into the sylvian fissure. A single dose of carboplatinum had no apparent effect on tumor size. Starting at age 38 months, he received 6 pulses, administered monthly, of bleomycin and vinblastine by constant infusion on 3 consecutive days, followed by thiotepa by constant infusion on 2 consecutive days. MRI Acquisition. Five sequential MRI scans were obtained for clinical purposes in the coronal plane on a 1.0 T Siemens Magnetom MR system (Iselin, NJ). Contiguous, thinly sliced three-dimensional TIweighted spoiled gradient echo FLASH sequences were performed with and without gadolinium-DTPA enhancement, with the following parameters: TR 40 msec, TE 15 msec, flip angle 50°, field of view 30 cm, slice thickness 3.1 mm, matrix 256 x 256 × 63, with 1 average and an imaging time of 11 min [20]. After the infusion of gadolinium and before beginning the Tl-weighted FLASH post-infusion (PI) sequence, a routine T2-weighted acquisition was performed with the following parameters: TR 2,500 msec, TE 35 msec, field of view 20 cm, slice thickness 5 mm with a 1.5 mm interslice gap, matrix 256 × 256, with 4 averages, and an imaging and reconstruction time of 12-13 min.

Communications should be addressed to: Dr. Filipek; Center for Morphometric Analysis, Neuroscience Center; MGH-East; 149 13th Street; Charlestown, MA 02129. Received January 29, 1991; accepted March 26, 1991.

Filipek et al: MRI-based Volumes of CNS Neoplasm

347

The pathophysiology of gadolinium enhancement in tumors is not well known. It has been shown that the degree of gadolinium enhancement in many tumors reaches a plateau in sequences begun between 15 (with a sequence mid-point of 18.5 rain) and 40 min PI 121,22]. In this study, the 11 min Ti-weighted PI sequences were begun approxinmtely 12-13 min after the infusion of gadolinium, giving a uniforrn midscan time of approximately 18-19 min P1 across the 5 sequential .scans. The first MRI scan of this series was performed on the first day of the second chemotherapy pulse. The second through fifth scans were obtained 19, 90, 124, and 187 days, respectively, after the first scan. Visual Interpretation. The radiologic reports of these 5 MRI scans from the distant imaging site were reviewed. The first 2 scans were interpreted by both a neuroradiology fellow and staff member, while the last 3 were read by a staff neuroradiologist alone. An interval change in the size of the mass was not appreciated when each of the first 4 scans was compared with the immediate preceding scan. On the fifth scan, an increase in the size of the mass was observed when compared with the fourth scan, with a new anterior extension of the enhancing mass. Volumetric Analysis. The MRI data sets were transferred by magnetic tape to our site for volumetric analyses. The image data sets were processed on a Sun Microsystems, Inc. (Mountain View, CA) #4/280 computer workstation. Each MRI scan required less than 30 min of investigator time for volumetric analysis. On each planar MRI, segmentation of the mass was performed by a single investigator using the intensity contour mapping and differential intensity contour algorithms on the Ti-weighted PI FLASH sequences [20,23-25]. These algorithms segment regions of interest based on the classification and identification of only the voxel locations constituting a border or surface of a given structure [23], rather than using a thresholding approach. Based on signal intensity histograms of the bright gadolinium enhancement and the surrounding tissue, an intensity level was defined for the border of the mass as the nadir of this bimodal distribution. The morphometric algorithms created a continuous outline around the mass based on the absolute or differential voxel (volume picture elements) signal intensity values. These operations have previously been described in greater detail [20,23-25]. Volumetric measures from the T2-weighted sequences and correlation with the Ti-weighted findings were not considered feasible, as prior experience points to an excessive error for volumetric analyses of such a small region on thicker skip-serial slices. The error of this morphometric method, principally due to volume averaging effects and segmentation procedures, has been previously estimated to be approximately 5% for large "hemispheric" structures, based on phantom studies [20]. With a fixed MR slice thickness, the error is assumed to increase with smaller or irregular regions of interest due to the greater partial volume contributions [20,23]. Preliminary analyses on smaller structures demonstrate a potential error of this method of less than 10% [26]. The volume of the right cerebellar hemisphere, the only unaffected and therefore presumably invariant structure, was computed from each scan, to assess the systematic variability of the method. The coefficient of variation (CV) [27] was determined for these cerebellar volumes. Other smaller structures may have been potentially more suitable fur comparison, but were distorted in both hemispheres by postsurgical changes or mass effect. The segmentation method creates contours at interfaces between regions of differing signal intensities. Although the cerebellar hemisphere and lesion have different signal intensities, the principle of border detection for the mass relative to sunxmnding brain is the same as for the cerebeUar hemisphere relative to surrounding structures or cerebrospinal fluid. Head alignment and, therefore, the apparent orientation of the mass in the image planes varied from scan to scan. All 5 MRI scans and segmentation outlines were subjected to a positional normalization protocol [25,26,28] (and unpublished data) which permitted more accurate definition of the borders of the mass on uniform slices across the sequential scans. In phantom studies, the error rate doubled for volumetric measurements using a combination of CT gantry angles [29]. It also was demonstrated that nonuniformity of the MRI plane

348

PEDIATRIC NEUROLOGY

Vol. 7 No. 5

orientation may impose a C V as high as l(/<,f acros~ ,alytn~ angfe3 t~, smaller structures similar to the volumes reported herc [26! The halving and doubling times were calculated durin,: ezlch mtcr scan interval, using the fbllowing equation II 31: T* t (In 2J Illa (virgil l

where V~) and V are the volumes at interval tinsel and t'll(t, respectively: T* is the doubling or halving time. and t is tilt: interval time itl days.

Results T h e m a s s w a s l o c a t e d in t h e b a s a l f o r e b r a i n o f t h e left hemisphere,

a d j a c e n t to t h e i n f e r o - l a t e r a l a s p e c t o f t h e

l a t e r a l v e n t r i c l e . It e x t e n d e d a l o n g t h e i n f e r i o r m a r g i n o f the ventricle from the level of the genu of the corpus c a l l o s u m r o s t r a l l y to t h e a m y g d a l a c a u d a l l y . D o r s o l a t e r a l ly it e x t e n d e d i n t o a n d l a r g e l y r e p l a c e d t h e g l o b u s p a t l i d u s a n d c a u d a t e n u c l e i . M e d i a l l y it i n v a d e d t h e i n t e r n a l c a p sule and hypothalamus.

As visualized on MRI, the mass

was represented primarily by a uniformly gadolinium-enhancing

compartment

( F i g 1). N o n e n h a n c i n g

cystic-ap-

pearing loculations were scattered throughout and adjacent to t h e e n h a n c i n g formed

separately

compartment

mass. Volumetric

analyses

on the gadolinium-DTPA

[GADO (+) REGIONS]

i n g l o c u l a t i o n s [ G A D O (-) R E G I O N S ]

were perenhancing

and the nonenhancas visualized on the

planar images. The GADO

volumetric

calculations

(-) c o m p a r t m e n t s

for the GADO

(+) and

of the mass, and the right cere-

b e l l a r h e m i s p h e r e a r e p r e s e n t e d i n T a b l e 1. T h e C V f o r t h e measured volume of the right cerebeUar hemisphere

was

Figure 1. Representative coronal slice from the first MRt depicting the choroid plexus carcinoma involving the left deep gray nuclear region adjacent to the lateral ventricle. The cystic-appearing loculations can be seen (arrows) within the gadolinium-enhancing mass.

Table 1.

Volume estimates of components of the mass Absolute Volume in cm 3 (scan number) 2 3 4

1

5

GADO (+) Region

6.46

4.85

2.01

0.94

2.30

GADO (-) Region

2.69

2.29

1.92

2.39

4.25

Total Mass

9.15

7.14

3.93

3,33

6.55

Cerebellum

70.65

72.70

72.75

73.20

74.81

1.83% for the 5 MRI scans which demonstrates the reproducibility of this method in accurately calculating structural or lesion volumes and suggests that the variations in the measured compartments reflected actual change in the mass rather than technical variability in the image acquisition or morphometric analyses. The interval percentage changes in volume for the G A D O (+) and total mass are listed in Table 2. Using a uniform spherical model, the change in the calculated radius of each of these compartments in each interval is also observed. The total volume of the mass decreased progressively during the 6-month course of chemotherapy. The volume then increased considerably by the fifth scan which was performed 63 days after the last pulse of chemotherapy (Fig 2). The decrease in volume reflected primarily a decline in the G A D O (+) region which decreased by 85.4% during the 6-month chemotherapy regimen (6.46 to 0.94 cm3; residual volume 14.6% of original). The halving times of the G A D O (+) region, 46 and 56 days respectively during the first 2 interscan intervals, accelerated to 31 days in the interval between scans 3 and 4. The volume of the multiple nonenhancing G A D O (-) loculations, in contrast to that of the G A D O (+) region, remained relatively stable throughout the initial period of observation. Both the G A D O (+) and G A D O (-) regions contributed to the surge in volume after the completion of chemotherapy, during the interval between scans 4 and 5. In this final interscan interval, the G A D O (+) region doubling time was 49 days, while that of the total mass was 65 days. The degrees of angulation varied among all 5 scans. The angle of rotation in the axial plane spanned 7.17 °, ranging Table 2.

from -5.97 to +1.20 °. The angle of rotation in the sagittal plane ranged from 0 to +14.49 °. The angle of rotation in the coronal plane spanned 4.16 °, ranging from -1.23 to +2.93 ° .

Discussion

Our report demonstrates that morphometry is a sensitive means of estimating change in vivo in the volume of an intracranial mass which was not appreciated from visualizing the two-dimensional images. Volumetric decreases of up to 59% were observed only when quantitative methods were used; however, as presented in Table 2, these volumetric changes were accompanied by minimal changes in the calculated radii of the G A D O (+) or total mass components. These calculated radii, based on a uniform spherical model, correspond to a maximal change across the 5 scans of less than 3 pixels. The actual in vivo changes in the radius would be even more difficult to recognize due to the following: (1) the configuration of the enhancing and nonenhancing components of the mass being highly irregular, and (2) the degrees of angulation among the 5 scans being most pronounced in the sagittal plane, producing significant variability of the presentation of the mass in the anterior-to-posterior extent. Therefore, visual evaluation was most likely compromised because of the arduous task of determining clinically significant change in three-dimensional volume based on changes in width or area of enhancing and nonenhancing regions as visualized on contiguous planar images. These measurements estimate the volume of the mass as demonstrated by signal intensity variations within the im-

A Volume and A calculated radius Interscan Interval

1~

2-3

3--4

4-5

-24.92%

-58.56%

-53.24%

+ 144.68%

GADO (+) Region

A Volume A Radius

-1.1 mm

-2.7 mm

-1.8 mm

+2.1 mm

Total Mass

A Volume A Radius

-21.97% -1.1 mm

-44.96% -2.2 mm

- 15.27% ~3.6 mm

+96.70% +2.4 mm

Filipek et al: MR/-based Volumesof CNS Neoplasm 349

VOLUME CHANGE WITH TIME AS VISUALIZED ON MRI VOLUMEIN CUBICCENTIMETERS

jfJ

1

2

3

4

5

6

7

MONTHS OF STUDY GADO (+) REGION

- 0 - GADO (-) REGION

~ " TOTAL MASS

Figure 2. Graphic representation of volumetric estimates for each component of the mass. Note that, although the GADO (+) region and TOTAL MASS consistently decreased over the first 4 scans, the GADO (-) region remained similar or increased, reflecting the addition of probable necrotic tumor to this compartment, and~or stabilization of the blood-brain barrier. Arrowheads represent timing of chemotherapy pulses relative to the MRI scans.

ages. These variations are, in actuality, measurements of the degree and extent of the blood-brain barrier breakdown rather than direct measures of histopathologically-concordant tumor mass. For many tumor types, the zone of contrast enhancement in CT usefully approximates but does not correlate precisely with the extent of histopathologically defined tumor [10,30-32], which is anticipated to be similar to gadolinium-DTPA in MRI. Correlation may be less reliable when the surrounding tissue is affected by enhancing postsurgical or radiation-induced changes. Radiation therapy was not employed in our study. Although surgical rim enhancement can persist for many years [3,33], the original areas of enhancement had disappeared prior to the initiation of this sequential volumetric study. Sequential quantitative analyses based on MRI allow additional in vivo observations of tumor behavior otherwise not obtainable. The rate of change in tumor volume, as measured by the doubling or halving time, is known to be a sensitive parameter for predicting therapeutic response and the probability of tumor recurrence following treatment [13-19]. These parameters can be accurately estimated from sequential MRI-based volumetric analyses. As an example, the rate of volume change for this choroid plexus carcinoma as visualized on MRI can be seen to differ with each interscan interval (slope of lines, Fig 2). During the 6 months of chemotherapy, the halving times of the GADO (+) region sequentially changed from 46-56 days, then to 31 days during the final Irealment interval. The complexity of the behavior of this mass is also reflected in the estimate of 49 days doubling time for the

350

PEDIATRIC NEUROLOGY

Vol. 7 No. 5

GADO (+) region on the fifth scan which is assumed to represent proliferating tumor as visualized on MRI. This estimate is based on only a single post-therapy datum point. It should be considered to be a maximum value, overestimated to the extent that the final interscan interval represents presumed initial tumor kill and subsequent recurrent growth in the 63-day period following the final chemotherapy pulse; therefore, the actual doubling time may have been substantially shorter, suggesting that this choroid plexus carcinoma may have been among the most rapidly growing tumors [17]. It is likely that the change of volume of the mass with time is the result of multiple incremental and decremental processes, including necrosis and dissolution, but also unconstrained growth of unresponsive cell populations. This complexity is implied by the differential patterns of volume change of the GADO (+) and GADO (-) regions. During the third interscan interval, where the rate of GADO (+) volumetric decrease is the greatest, there is a reciprocal relationship of the contributions made by these two compartments of the total visualized mass. Presumably this phenomenon reflects, at least in part, a shift of necrotic tissue from the GADO (+) into the nonenhancing GADO (-) compartment. The potential for three-dimensional correlation of these quantitative morphometric analyses with studies of tumor physiology (e.g., positron emission tomography, MRI spectroscopy, MRI perfusion) will be critical to the development of a complete understanding of the biologic significance of these observations. These coordinated physio-

logic and anatomic analyses have broad implications for the in vivo study of brain tumor biology and its treatment. Volumetric measurements can be performed efficiently and with sufficient accuracy to identify clinically subtle changes in CNS tumor volumes with an acceptably low rate of variability which permit in vivo observations of their biologic properties. These analyses can be performed with relatively little effort beyond the routine acquisition of MRI, but have the capacity to expand the understanding of tumor kinetics and response to therapeutic intervention. If one extrapolates these observations to larger clinical trials, one could predict the ability to correlate initial growth rates with therapeutic regression for many tumor types which has implications for defining the natural history of tumors and the responsiveness of given tumor types to therapeutic intervention. This work was supported in part by grants NS 24279 and NS 20489 (Drs. Filipek and Kennedy) from the National Institute of Neurologic Disorders and Stroke and CA 40303 (Dr. Kennedy) from the National Cancer Institute, Bethesda, MD. The authors wish to express their gratitude to the following people who were indispensable for the successful completion of this study: Drs. Bruce R. Rosen, Keith R, Thulborn, David Mikulis, and Willem du Bois; Drs. Lawrence Helson, Ronald Jacobson, Jeffrey Allen, and Joseph Zito; and the staff of the MRI Facility at Long Island Jewish Medical Center. References

[1] Schroth G, Grodd W, Guhl L, Grauer M, Klose U, Niendorf HP. Magnetic resonance imaging in small lesions of the central nervous system. Improvement by gadolinium-DTPA. Acta Radiol 1987;28: 667-72. [2] Graif M, Steiner RE. Contrast-enhanced magnetic resonance imaging of tumors of the central nervous system: A clinical review. Br J Radiol 1986;59:865-73. [3] Cohen BH, Bury E, Packer RJ, Sutton LN, Bilaniuk LT, Zimmerman RA. Gadolinium-DTPA-enhanced magnetic resonance imaging in childhood brain tumors. Neurology 1989;39:1178-83. [4] Stack JP, Antoun NM, Jenkins JP, Metcalfe R, Isherwood I. Gadolinium-DTPA as a contrast agent in magnetic resonance imaging of the brain. Neuroradiology 1988;30:145-54. [5] Valk J. Gd-DTPA in MR of spinal lesions. AJR 1988;150: 1163-8. [6] Ross JS, Masaryk T J, Modic MR. Three-dimensional FLASH imaging: Applications with gadolinium-DTPA. J Comput Assist Tomogr 1989; 13:547-52. [7] Albright RE, Fram EK. Microcomputer-based technique for 3-D reconstruction and volume measurement of computed tomographic images. Part 2: Anaplastic primary brain tumors. Invest Radiol 1988; 23:886-90. [8] Criscuolo GR, Oldfield EH. Measurement of intracranial tissue volume using computed tomographic images and a personal computer. Neurosurgery 1988;23:671-4. [9] Friedman MA, Resser KJ, Marcus RS, Moss~RA, Cann CE. How accurate are computed tomographic scans in assessment of changes in tumor size? Am J Med 1983;75:193-8.

[10] Levin VA, Hoffman WF, Heilbron DC, Norman D. Prognostic significance of the pretreatment CT scan on time to progression for patients with malignant gliomas. J Neurosurg 1980;52:642-7. [11] Yamashita T, Kuwabara T. Estimation of rate of growth of malignant brain tumors by computed tomographic scanning. Surg Neurol 1983;20:464-70. [12] Laasonen EM, Troupp H. Volume growth rate of acoustic neurinomas. Neuroradiology 1986;28:203-7. [13] Schwartz M. A biomathematical approach to clinical tumor growth. Cancer 1961; 14:1272-94. [14] Steel GG, Lamberton LE The growth rate of human tumors. Br J Cancer 1966;20:74-86. [15] Steel GG. Cell loss as a factor in the growth rate of human tumors. Eur J Cancer 1967;3:381-7. [16] Steel GG. The growth kinetics of tumors in relation to their therapeutic response. Laryngoscope 1975;85:359-70. [17] Shackney SE, McCormack GW, Cuchural GJ. Growth rate patterns of solid tumors and their relation to responsiveness to therapy. Ann Intern Med 1978;89:107-21. [18] Tubiana M. The growth and progression of human tumors: Implications for management strategy. Radiother Oncol 1986;6:167-84. [19] Tubiana M. Tumor cell proliferation kinetics and tumor growth rate. Acta Oncol 1989;28:113-21. [20] Filipek PA, K e n n e d y DN, Caviness VS, Rossnick SL, Spraggins TA, Starewicz PM. Magnetic resonance imaging-based brain morphometry: Development and application to normal subjects. Ann Neurol 1989;25:61-7. [21] Schorner W, Laniado M, Niendorf HP, Schubert C, Felix R. Time-dependent changes in image contrast in brain tumors after gadolinium-DTPA. AJNR 1986;7:1013-20. [22] Stehling MK, Bullock P, Firth JL, et al. Gd-DTPA real-time studies of the brain with EPI: A dynamic approach to perfusion and blood-brain-barrier assessment. Proc Soc Mag Res Med 1989;8:358. [23] Kennedy DN, Filipek PA, Caviness VS. Anatomic segmentation and volumetric calculations in nuclear magnetic resonance imaging. IEEE Trans Med lmag 1989;8:1-7. [24] Caviness VS, Filipek PA, Kennedy DN. Magnetic resonance technology in human brain science: A blue print for a program based upon morphometry. Brain Dev 1989; 11:1-13. [25] Filipek PA, Kennedy DN, Caviness VS. A method of morphometric analysis of the human brain based upon magnetic resonance imaging. Ann Neurol 1988;24:356. [26] Filipek PA, Kennedy DN, Rademacher J, Caviness VS. Error and variability incurred with MRI-based morphometry. Ann Neurol 1990;28:459. [27] Colton T. Descriptive statistics. In: Statistics in medicine. Boston: Little, Brown and Company, 1974; 11-61. [28] Filipek PA, Kennedy DN, Caviness VS. Morphometric analysis of central nervous system neoplasms. Ann Neurol 1989;26:461. [29] Albright RE, Fram EK. Microcomputer-based technique for 3-D reconstruction and volume measurement of computed tomographic images. Part h Phantom studies. Invest Radiol 1988;23:881-5. [30] Burger PC. Pathologic anatomy and CT correlations in glioblastoma multiforme. Appl Neurophysiol 1983;46:180-7. [31] Johnson PC, Hunt SK, Drayer BP. Human cerebral gliomas: Correlation of postmortem MR imaging and neuropathologic findings. Radiology 1989;170:211-7. [321 Earnest F, Kelly PJ, Scheithauser BW, et al. Cerebral astrocytomas: Histopathological correlation of MR and CT contrast enhancement with stereotactic biopsy. Radiology 1988; 166:823-7. [33] Cairneross J, Pexman JHW, Rathbone MP. Post-surgical contrast enhancement mimicking residual brain tumor. Can J Neurol Sci 1985;12:75.

Filipek et al: MRI-based Volumes of CNS Neoplasm 351