Department of Neurosurgery, The Cleveland Clinic Foundation, Cleveland, Ohio

Transcription

1 Int. J. Cancer (Radiat. Oncol. Invest): 96, (2001) 2001 Wiley-Liss, Inc. Publication of the International Union Against Cancer The Sensitivity and Specificity of FDG PET in Distinguishing Recurrent Brain Tumor from Radionecrosis in Patients Treated with Stereotactic Radiosurgery Samuel T. Chao, 1 John H. Suh, M.D., 1 * Shanker Raja, M.D., 2 Shih-Yuan Lee, M.S.P.H., 1 and Gene Barnett, M.D. 3 1 Department of Radiation Oncology, The Cleveland Clinic Foundation, Cleveland, Ohio 2 Department of Nuclear Medicine, The Cleveland Clinic Foundation, Cleveland, Ohio 3 Department of Neurosurgery, The Cleveland Clinic Foundation, Cleveland, Ohio SUMMARY Radiation necrosis and recurrent brain tumor have similar symptoms and are indistinguishable on both magnetic resonance imaging (MRI) and computed tomograph scans. 18 F-fluorodeoxyglucose (FDG) positron emission tomography (PET) has been proposed as a diagnostic alternative, particularly when co-registered with MRI. We studied 47 patients with brain tumors treated with stereotactic radiosurgery and followed with FDG PET. For all tumor types, the sensitivity of FDG PET for diagnosing tumor was 75% and the specificity was 81%. For brain metastasis without MRI co-registration, FDG PET had a sensitivity of 65% and a specificity of 80%. For brain metastasis with MRI co-registration, FDG PET had a sensitivity of 86% and specificity of 80%. MRI coregistration appears to improve the sensitivity of FDG PET, making it a useful modality to distinguish between radiation necrosis and recurrent brain metastasis Wiley-Liss, Inc. Key words: positron emission tomography; radionecrosis; stereotactic radiosurgery; brain tumor INTRODUCTION Radionecrosis of brain tumors has become more frequent with the advent of aggressive radiation treatment options such as brachytherapy and stereotactic radiosurgery (SRS). Radionecrosis may arise months to years after treatment and can lead to headache, weakness, seizures, focal neurological deficits, and increased intracranial pressure [1]. Unfortunately, these symptoms are similar to recurrence of the brain tumor. Distinguishing between recurrence and radionecrosis is essential in directing therapy. Radiation necrosis is usually treated with steroids, with surgical resection being reserved for neurological deterioration or steroid dependency. The choice of treatment of recurrent tumor depends on the patient s treatment history; options include steroids, surgery, radiotherapy, and stereotactic radiosurgery. Biopsy of the lesion is the gold standard but is invasive and expensive. Neither computed tomography (CT) nor magnetic resonance imaging (MRI) scans can distinguish between recurrent tumor and radionecrosis because both lesions demonstrate mass effect and contrast enhancement [2 4]. Other non-invasive modalities that have been used to differentiate the two are positron emission tomography (PET), single photon emission computerized tomography (SPECT), and magnetic resonance (MR) spectroscopy. For PET, it is necessary to use metabolic trac- *Correspondence to: John H. Suh, M.D., Department of Radiation Oncology, Desk T-28, The Cleveland Clinic Foundation, 9500 Euclid Ave., Cleveland, OH Phone: (216) ; Fax: (216) ; suh@radonc.ccf.org Received 19 January 2001; Revised 9 March 2001; Accepted 14 March 2001 Published online 19 April 2001

2 192 Chao et al.: FDG PET for Distinguishing Recurrent Brain Tumor from Radionecrosis ers such as 18 F-fluorodeoxyglucose (FDG), 11 C- methionine, or 131 I-iododeoxyuridine [5]. FDG is the most widely used tracer for studying brain lesion metabolism. It is taken up by cells and phosphorylated by hexokinase to FDG-6-PO 4, which is reverted to FDG by glucose-6-phosphatase. Tumor cells accumulate FDG because they are hypermetabolic (with an increased rate of glycolysis) with elevated hexokinase activity and lower glucose-6- phosphatase activity [6]. In contrast, radiation necrosis is hypometabolic, and thus, FDG should not accumulate in necrotic tissue. The FDG level can be measured by PET cameras when the excess proton in 18 F decays and produces, among other particles, a positron that interacts with an electron and emits two photons 180 degrees apart [7,8]. The use of FDG PET to distinguish radiation necrosis from recurrent tumor in the brain has been studied with mixed results. Some institutions report high accuracy [9 13], whereas others report that PET is neither sensitive enough nor specific enough to be used routinely [7,14 16]. MRI co-registration, which fuses MRI images with PET images, has been proposed as a way to improve the accuracy of PET [17,18]. This procedure, which provides both morphological and physiological information about a lesion, has been used at our institution since In our retrospective study, we calculated the sensitivity and specificity of PET in distinguishing recurrent brain tumors from radionecrosis in 47 patients with previous brain tumors treated with SRS. We also investigated whether MRI co-registration improved the accuracy of PET. MATERIALS AND METHODS Table 1. Diagnoses in 47 Patients with Brain Lesions Followed by FDG PET Diagnosis Number of patients Metastases (n 32) Lung (nonsmall-cell) metastasis 19 Lung (small cell) metastasis 1 Breast metastasis 3 Renal cell metastasis 4 Melanoma metastasis 1 Head and neck metastasis 1 Endometrial metastasis 1 Unknown metastasis 2 Primary brain tumors (n 15) Glioblastoma multiforme 10 Anaplastic astrocytoma 3 Pilocytic astrocytoma 1 Low-grade astrocytoma 1 FDG 18 F-fluorodeoxyglucose; PET positron emission tomography. Patient Sample We reviewed the records of patients who had undergone either linear accelerator-based (LINAC) or Leksell Gamma Knife stereotactic radiosurgery between August 7, 1990, and July 14, 1999, for a primary brain tumor or brain metastasis. According to the available records, 47 of these patients had an FDG PET scan during the course of their illness. In 32 of these patients, the original diagnosis was brain metastasis, and in the remaining 15, primary brain tumor (Table 1). Twenty-five patients were male and 22 were female. The median age was 57 years (range 6 75 years). The total number of lesions was 88. Twenty-four patients were treated with Gamma Knife, 22 with LINAC radiosurgery, and 1 with both. Patients were followed up with MRI, CT, or both. Because of radiographic features suggestive of recurrence or radiation necrosis, the patients underwent further evaluation with PET scan. Ten patients had repeat PET scans. The median interval between SRS and PET was 6.1 months (range months). Stereotactic Radiosurgery LINAC radiosurgery was used for all 23 patients treated before December Doses ranged from 1,000 to 2,400 cgy delivered to the 50% to 80% isodose line. The 24 remaining patients treated after December 1996 were all treated with the Leksell Gamma Knife, as was one of the LINAC patients who needed a second treatment. These patients received 1,000 to 2,400 cgy delivered to the 40% to 95% isodose line. FDG PET Imaging All patients were scanned on a Posicam (Positron Corp., Houston, Tex) PET scanner 45 to 60 minutes after receiving 5 to 10 mci ( MBq) of 18 FDG intravenously. Projection data were obtained for 20 minutes or 20 million counts. The projection data were reconstructed using accepted back-projection algorithms with Butterworth filter. The reconstructed images were reviewed on a continuous color scale (gray scale) on a monitor. Any area with more uptake than the adjacent gray/white matter was considered suspicious for tumor. Additionally, any uptake of FDG in a site showing contrast enhancement on coregistered MRI was considered abnormal and indicative of recurrent tumor. All lesions showing no enhanced uptake of 18 F were considered radiation necrosis. To establish the sensitivity and specificity of

3 Chao et al.: FDG PET for Distinguishing Recurrent Brain Tumor from Radionecrosis 193 Table 2. FDG PET Diagnoses Compared with a Gold Standard of Biopsy/Radiographic Follow-up in 44 Lesions FDG PET diagnosis Gold standard Tumor Necrosis Tumor Necrosis Tumor Necrosis Tumor Necrosis Sensitivity 75% 71% 86% Specificity 81% 80% 100% FDG 18 F-fluorodeoxyglucose; PET positron emission tomography. brain metastasis patients only (n 36) primary tumor patients only (n 8) FDG PET, we compared the FDG PET diagnosis with either the pathological diagnosis or long-term clinical and radiographic diagnosis. Pathological Confirmation Pathology samples were available for 18 lesions (5 that had biopsy alone, 8 that were treated with gross total resections, 2 that were treated with subtotal resections, 1 autopsy sample, and 2 lesions in which the degree of resection could not be determined because the records had been lost). In one case, pathology records were not obtainable because the resection had been done at an outside hospital. Radiographic Confirmation Thirty-four patients (62 lesions) had follow-up MRI, CT, or both. Median follow-up was 5.6 months (range months). A diagnosis of presumed radiation necrosis was made if the lesion decreased in size over long-term follow-up. Lesions re-treated with Gamma Knife, LINAC, or chemotherapy were characterized as tumor if the clinical and radiographic evidence was compelling. Lesions that had no change or increased in size were characterized as indeterminate. RESULTS Lesion diameter in this study on follow-up ranged from 0 to 4.5 cm, with a median diameter of 2 cm. Tables 2 and 3 compare FDG PET results with those of biopsy and radiographic follow-up. Overall, the FDG PET results identified 24 lesions as tumors, and 21 of these were true positives, as confirmed by histologic examination (n 11) or radiographic follow-up (n 10). The other three lesions were false positives, as confirmed by histologic examination (n 2) or radiographic follow-up (n 1). The FDG PET results ruled out tumor in 20 lesions, 13 of which were true negatives (all determined by radiographic follow-up). The other seven lesions were false negatives, as confirmed by histologic examination (n 4) or radiographic follow-up (n 3). Some patients (n 14) had MRI coregistration during analysis of their PET scans. Table 4 compares the results of FDG PET with MRI co-registration with those of biopsy and radiographic follow-up. The number of patients with FDG PET with MRI co-registration and biopsy alone was too small and the results were not reported. For brain metastasis alone, using biopsy or radiographic and clinical follow-up as the gold standard, the sensitivity and specificity was 86% and 80%, respectively. These values compare with a sensitivity of 65% and a specificity of 80% for FDG PET without MRI co-registration. DISCUSSION Currently, the gold standard for distinguishing tumor recurrence from radiation necrosis is biopsy, which has an accuracy and specificity of more than 95% [19,20]. However, biopsy is invasive and has numerous potential complications such as infection, neurological problems, and hematoma [21,22]. In contrast, FDG PET is non-invasive and safe with few contraindications. The US Health Care Financing Agency and some insurance companies have approved FDG PET for a variety of malignant conditions, although not for distinguishing between radiation necrosis and tumor recurrence in the brain. Thus, some patients would have to pay the estimated US$1, out of pocket. FDG PET is available at few sites because the FDG tracers are generated by cyclotron and shipped to the clinical site. Several studies have supported using FDG PET for differential diagnosis of radiation necrosis. In 1982, Patronas et al. [11] described five patients with astrocytomas who had undergone radiation therapy and subsequent FDG PET, which accurately identified two cases of radiation necrosis and three cases of recurrent tumor. The gold standard

4 194 Chao et al.: FDG PET for Distinguishing Recurrent Brain Tumor from Radionecrosis Table 3. FDG PET Diagnoses Compared with a Gold Standard of Biopsy Only in 17 Lesions FDG PET diagnosis brain metastasis patients only (n 13) primary tumor patients only (n 4) Gold standard Tumor Necrosis Tumor Necrosis Tumor Necrosis Tumor Necrosis Sensitivity 73% 73% 75% Specificity NA a FDG 18 F-fluorodeoxyglucose; PET positron emission tomography. a NA not available, because only two patients underwent biopsy. Table 4. Diagnoses with FDG PET with MRI Coregistration Against a Gold Standard of Biopsy/Radiographic Follow-up in 14 lesions FDG PET diagnosis brain metastasis patients only (n 12) primary tumor patients only (n 2) Gold standard Tumor Necrosis Tumor Necrosis Tumor Necrosis Tumor Necrosis Sensitivity 75% 86% 0% Specificity 83% 80% 100% FDG 18 F-fluorodeoxyglucose; MRI Magnetic resonance imaging; PET positron emission tomography. was biopsy or autopsy. More recently, Kim et al. [10] reported that FDG PET had a sensitivity of 80% and specificity of 94% in 33 cases of primary brain tumors and brain metastases. Of these, eight patients underwent biopsy and the rest were followed up clinically over 9 months. Other studies by Ericson et al. [9] of brain metastases (n 31) and Ogawa et al. [12] of primary brain tumors (n 15) also suggest that FDG PET may have an important role in distinguishing radiation necrosis from tumor recurrence. In contrast, several other studies suggest that FDG PET is neither sensitive enough nor specific enough to distinguish between radionecrosis and recurrent tumor. Kahn et al. [16] studied 19 patients with a variety of brain tumors and compared the results with those of Tl 201 SPECT. The sensitivity and specificity of FDG PET in that study were 81% and 40%, respectively. These results were not statistically significantly different from those of Tl 201 SPECT, and thus, the authors argued that the increased cost of FDG PET was not justified [16]. Griffeth et al. [14] used FDG PET to study 31 known brain metastases and found that FDG PET detected hypermetabolism in 68% of the lesions. Given that all tumors are theoretically hypermetabolic, this result means that FDG PET is only 68% sensitive in the detection of brain metastases [14]. Ricci et al. [15] studied 31 patients with history of glial neoplasms who underwent PET to determine whether new lesions were radiation necrosis or recurrence. The results were confirmed by biopsy. The sensitivity was between 73% and 86% and the specificity was between 22% and 56%, depending on the technique used. Given those results, up to one-third of the patients would have been treated inappropriately [15]. Recently, Thompson et al. [7] retrospectively reviewed the records of 15 patients who had PET and subsequent biopsy, and reported that the sensitivity was 43%. There was only one patient with radiation necrosis who was correctly picked up by FDG PET, for a specificity of 100% [7]. These results have led to controversy over the use of FDG PET to diagnose recurrence of a brain tumor. In our study, when biopsy alone was used as the gold standard, the sensitivity of FDG PET in detecting recurrent brain metastasis or primary brain tumor was 73%. When patients with brain metastases or glial tumors were considered as separate groups, the sensitivities were not much different, at 73% and 75%, respectively. Because our study was retrospective, patients with negative PET scans were less likely to undergo biopsy, thus making the sample size (n 2) too small to calculate specificity. We thus included supporting evidence from radiographic and clinical follow-up. When the gold standard was biopsy or radiographic and

5 Chao et al.: FDG PET for Distinguishing Recurrent Brain Tumor from Radionecrosis 195 clinical diagnosis, the sensitivity and specificity of PET became 75% and 81%, respectively. The accuracy was 77%. When patients with brain metastases or glial tumors were considered separately, the sensitivity of FDG PET was 71% and 86%, respectively. Specificity for brain metastasis alone was 80%. Specificity for glial tumors was calculated to be 100%, but only one glial tumor patient had evidence of radionecrosis. There were seven false negatives and three false positives of the 44 lesions that could be studied. One of these false negatives was a patient with history of a lung metastasis to the cerebellum who underwent biopsy after FDG PET. Intraoperatively, the diagnosis came back as radiation necrosis, but final pathology was positive for malignant cells. Based on the intraoperative diagnosis, this lesion would have been a true negative. However, because final pathology is more sensitive to picking up the rare microscopic malignant cells, the lesion was considered a false negative. Similarly, in another case, the FDG PET of a Gamma Knife-treated carcinoid metastasis to the right occipital lobe had features of tumor recurrence. Intraoperatively, the diagnosis was necrosis, which would have made this case a false positive, but on final pathological examination, tumor cells were found. Thus, this case was a true positive. This finding suggests that FDG PET is not effective at picking up the beginnings of tumor recurrence, thus making this test insensitive. False negatives may be dependent on the tumor size, the resolution of the PET scanner, and tumor histology. In one false-positive case, the initial PET of a metastatic lesion involving the mesial aspect of the right occipital lobe was read as negative. On repeat PET 78 days later, however, the patient had positive PET results. Subsequent biopsy was negative for recurrence, making this case a false positive based on the second FDG PET. False positives can be the result of the high FDG uptake in the normal cortex or local seizure activity, which can confound the interpretation of PET. Hustinx and Alavi [17] have suggested that MRI co-registration may help improve accuracy and decrease false positives. MRI co-registration has also been described by Thiel et al. [18] in a technical case report as a way to increase the accuracy of PET. MRI coregistration fuses high-resolution MRI scans with PET scans to better identify distinct regions of hypermetabolism and hypometabolism within the lesion. This method gives the physician both morphological and physiological information about the lesion. In addition to improving accuracy, this procedure may help improve treatment planning by both neurosurgeons and radiation oncologists. In our study, FDG PET with MRI coregistration, for brain metastasis patients alone, had a sensitivity of 86% and specificity of 80%, respectively. FDG PET without MRI co-registration had a sensitivity of 65% and a specificity of 80%. The sensitivity and specificity for glial tumors with PET and MRI co-registration had little precision given the small sample size of two lesions, and thus were not reported. Of note, in one of the cases, FDG PET was initially read as negative for recurrent tumor, but after MRI co-registration, the final read was positive for recurrent tumor. Thus, MRI coregistration seems to have improved the sensitivity of FDG PET from 65% to 86% for brain metastasis recurrence in SRS, thus making FDG PET a useful test in the diagnosis of recurrent brain metastasis. This study is limited by being a retrospective study. As discussed earlier, we had to supplement our biopsy results with those of radiographic and clinical follow-up to offset the fact that few patients with negative PET scans underwent biopsy. In addition, many patients were excluded because they lacked follow-up or follow-up was insufficient for a diagnosis. Thus, the number of lesions studied was low. However, most studies to date have been retrospective and were subject to the same complications encountered in the present study. Some studies have also been limited by small patient numbers. A meta-analysis of data from all related studies would be useful in generating more accurate sensitivity and specificity data. Ideally, a larger study using biopsy as the gold standard would help to better clarify the role of PET in diagnosing recurrent tumor and radiation necrosis. According to our results, FDG PET without MRI co-registration is only moderately effective in distinguishing between radiation necrosis and recurrent tumor. MRI co-registration appears to improve the sensitivity in brain metastasis patients, making this modality useful. A larger study of PET with MRI co-registration would be useful in verifying that this technique improves the sensitivity and specificity over PET alone. Other modalities such as L-methyl- 11 C- methionine PET (Met-PET) and MR spectroscopy may be useful as well. In Met-PET, L-methyl- 11 C- methionine is taken up by tumor cells by a mechanism that remains unclear. It is a good tracer because uptake in normal brain parenchyma, including cortex, is low. In a case report, Ogawa et al. [12] found that Met-PET correctly identified three cases of radiation necrosis and seven cases of

6 196 Chao et al.: FDG PET for Distinguishing Recurrent Brain Tumor from Radionecrosis recurrent tumor. Further studies are necessary to determine if Met-PET would be a useful modality in diagnosing recurrent brain tumor. MR spectroscopy also has shown promise, because in normal brain, it shows metabolic peaks from cholinecontaining compounds (CHO), creatine plus phosphocreatine (CR), and N-acetyl-containing compounds (NA). In tumor cells, CHO and CR are elevated, but NA is reduced. In necrosis, CHO, CR, and NA are reduced. MR spectroscopy correctly identified five of seven cases of active tumor and four of five cases of cerebral necrosis [23]. Again, a larger study is necessary to determine the role of MR spectroscopy in distinguishing recurrent tumor versus radiation necrosis. CONCLUSION Our data suggest that FDG PET alone is insensitive in distinguishing recurrent tumor from radiation necrosis, but MRI co-registration makes FDG PET a more sensitive and useful test in distinguishing brain metastasis recurrence from radionecrosis. The Cleveland Clinic has been using MRI coregistration with PET since 1997, and the data support continued use. We will continue to follow patients with FDG PET with MRI co-registration. ACKNOWLEDGEMENT We would like to thank Jessica Ancker for her help in preparing this manuscript. REFERENCES 1. Valk PE, Dillon WP. Radiation injury of the brain. AJNR Am J Neuroradiol 1991;12: Graeb DA, Steinbok P, Robertson WD. Transient early computed tomographic changes mimicking tumor progression after brain tumor irradiation. Radiology 1982; 144: Dooms GC, Hecht S, Brant-Zawadzki M, Berthiaume Y, Norman D, Newton TH. Brain radiation lesions: MR imaging. Radiology 1986;158: Ashdown BC, Boyko OB, Uglietta JP, Friedman HS, Hockenberger B, Oakes WJ, Fuller GN. Postradiation cerebellar necrosis mimicking tumor: MR appearance. J Comput Assist Tomogr 1993;17: Blasberg RG. Prediction of brain tumor therapy response by PET. J Neurooncol 1994;22: Hawkins RA, Choi Y, Huang SC, Messa C, Hoh CL, Phelps ME. Quantitating tumor glucose metabolism with FDG and PET. J Nucl Med 1992;33: Thompson TP, Lunsford LD, Kondziolka D. Distinguishing recurrent tumor and radiation necrosis with positron emission tomography versus stereotactic biopsy. Stereotactic Funct Neurosurg 1999;73: Xiong J, Nickerson LDG, Downs JF, Fox PT. Basic principles and neurosurgical applications of positron emission tomography. Neurosurg Clin 1997;8: Ericson K, Kihlstrom L, Mogard J, Karlsson B, Lundquist C, Widen L Collins VP, Stone-Elander S. Positron emission tomography using 18 F-fluorodoxyglucose in patients with stereotactically irradiated brain metastases. Stereotactic Funct Neurosurg 1996; 66: Kim E, Chung S, Haynie TP, Kim CG, Cho BJ, Podoloff DA, Tilbury RS, Yang DJ, Yung WK, Moser RP, Ajani JA. Differentiation of residual or recurrent tumors from post-treatment changes with F-18 FDG PET. Radiographics 1992;12: Petronas NJ, DiChiro G, Brooks RA, DeLaPaz RL, Kornblith PL, Smith BH, Rizzoli HV, Kessler RM, Manning RG, Channing M, Wolf AP, O Connor CM. Work in progress: [ 18 F] fluorodeoxyglucose and positron emission tomography in the evaluation of radiation necrosis of the brain. Radiology 1982;144: Ogawa T, Kanno I, Shishido F, Inugami A, Higano S, Fujita H, Murakami M, Uemura K, Yasui N, Mineura K, Kowada M. Clinical value of PET with 18 F- fluorodeoxyglucose and L-methyl- 11 C-methionine for the diagnosis of recurrent brain tumor and radiation injury. Acta Radiol 1991;32: Glantz MJ, Hoffman JM, Coleman RE, Friedman AH, Hanson MW, Burger PC, Herndon JE, Meisler WJ, Schold SC. Identification of early recurrence of primary central nervous system tumors by [ 18 F] fluorodeoxyglucose positron emission tomography. Ann Neurol 1991;29: Griffeth LK, Rich KM, Dehdashti F, Simpson JR, Fusselman MJ, McGuire AH, Siegel BA. Brain metastases from non-central nervous system tumors: evaluation with PET. Radiology 1993;186: Ricci PE, Karis JP, Heiserman JE, Fram EK, Bice AN, Drayer BP. Differentiating recurrent tumor from radiation necrosis: time for re-evaluation of positron emission tomography? AJNR Am J Neuroradiol 1998;19: Kahn D, Follett KA, Bushnell DL, Nathan MA, Piper JG, Madsen M, Kirchner PT. Diagnosis of recurrent 201 brain tumor: value of Tl SPECT vs 18 F- fluorodeoxyglucose PET. AJR Am J Roentgenol 1994; 163: Hustinx R, Alavi A. SPECT and PET imaging of brain tumors. Neuroimaging Clin N Am 1999;9: Thiel A, Pietrzyk U, Sturm V, Herholz K, Hovels M, Schroder R. Enhanced accuracy in differential diagnosis of radiation necrosis by positron emission tomography-magnetic resonance imaging co-registration: technical case report. Neurosurgery 2000;46: Chin LS, Levy ML, Rabb CH, Chandrasoma PT, Zee CS, Apuzzo MJ. Principles and pitfalls of image directed stereotactic biopsy of brain lesions. In: Thomas DGT, editor. Stereotactic and image directed surgery of

7 Chao et al.: FDG PET for Distinguishing Recurrent Brain Tumor from Radionecrosis 197 brain tumours. New York: Churchill Livingstone; p Lunsford LD. Diagnosis and treatment of mass lesions using the Leksell stereotactic system. In: Lunsford LD, editor. Modern stereotactic neurosurgery. Boston: Nijhoff; p Apuzzo ML, Sabshin JK. Computed tomographic guidance stereotaxis in the management of intracranial mass lesions. Neurosurgery 1983;12: Kelly PJ. Tumor stereotaxis. Philadelphia: WB Saunders; Taylor JS, Langston JW, Reddick WE, Kingsley PB, Ogg RJ, Pui MH, Kun LE, Jenkins JJ, Chen G, Ochs JJ, Sanford RA, Heideman RL. Clinical value of proton magnetic resonance spectroscopy for differentiating recurrent or residual brain tumor from delayed cerebral necrosis. Int J Radic Oncol Biol Phys 1996;36: