Volumetric Analysis of ¹⁸F-FDG PET in Glioblastoma Multiforme: Prognostic Information and Possible Role in Definition of Target Volumes in Radiation Dose Escalation

Study Design and Treatment

Patients were considered eligible to participate in the study if they had a diagnosis of GBM, were at least 18 y old, and had a score of at least 60 on the Karnofsky Scale. In a thermoplastic head immobilization mask, patients underwent CT for 3-dimensional computerized radiation treatment planning. Patient positioning was the same for CT as for postoperative MRI. For patients on whom postoperative MRI was performed at a University of Washington institution, MR and CT images were fused within the computerized system, called Prism, that the University of Washington has developed for radiation treatment planning. The initial planning target volume was defined as the T2-weighted MRI signal abnormality plus a 2.5- to 3-cm margin. The initial planning target volume was treated to a dose of 50.4 Gy (1.8 Gy per fraction), usually through 3 noncoplanar fields designed by the beams-eye-view tool of Prism. Fields were then reduced to the second planning target volume, designated as the T2-weighted MRI abnormality plus a 1.5-cm margin. An additional 9 Gy (1.8 Gy per fraction) was delivered to this second planning target volume, bringing the total initial dose to the standard 59.4 Gy. At a dose of between 45 and 50.4 Gy, patients underwent ¹⁸F-FDG PET. Because the operative bed anatomy may have changed over the course of radiation, an additional MRI study was performed at the time of the ¹⁸F-FDG PET scan for correlation with the ¹⁸F-FDG PET images.

All ¹⁸F-FDG PET studies were performed on an Advance tomograph (General Electric Medical Systems, Waukesha, WI). Images were acquired in 3 dimensions onto 35 × 128 × 128 matrices using 4-mm transverse and 8.5-mm axial filters. The result was images with 4- to 6-mm spatial resolution in both axial and transverse directions (16). ¹⁸F-FDG was prepared using the method of Hamacher et al. (17) and had radiochemical purity in excess of 95% and specific activity greater than 47 GBq/mmol in all cases. The patients fasted for at least 6 h before the PET study. After plasma glucose measurement to rule out hyperglycemia (glucose less than 150 mg/dL), 259–370 MBq of ¹⁸F-FDG were infused over 1–2 min. Starting at approximately 45 min after injection, a 15-min 3-dimensional emission acquisition was obtained, followed by a transmission study 25 min after injection (16). Images were acquired with the patients in their radiotherapy head immobilization mask to facilitate coregistration of the PET scans with the treatment-planning CT scans. Images were reconstructed using the manufacturer’s implementation of the 3-dimensional reprojection algorithm (18). Images were reconstructed in a 30-cm transverse plane by a 14.5-cm axial image field of view onto 128 × 128 × 35 matrices using 4-mm transverse and 8.5-mm axial filters. The resulting image spatial resolution was between 5.5 and 6.5 in full width at half maximum in both axial and transverse directions.

Identification of ¹⁸F-FDG-avid areas of tumor and delineation of the ¹⁸F-FDG PET-positive contours took place in 2 steps. First, the ¹⁸F-FDG-avid tumor areas were determined using MRI as an adjunct to ¹⁸F-FDG PET interpretation. Second, ¹⁸F-FDG PET images were coregistered to the radiotherapy-planning CT scans, and ¹⁸F-FDG-avid tumor contours were drawn on the reformatted ¹⁸F-FDG scans. Because patients underwent both PET and CT in the radiotherapy immobilization mask, which firmly constrains head positioning, the reformatted ¹⁸F-FDG PET scans used for treatment planning differed only in transverse slice width from the original PET images.

MRI performed close to the time of ¹⁸F-FDG PET was used to aid in the interpretation of the ¹⁸F-FDG PET studies. When MR images were obtained at the University of Washington and were therefore available in electronic form, the MR and ¹⁸F-FDG PET images were software coregistered to aid image interpretation. For MR images obtained elsewhere and available as hard-copy films only, the ¹⁸F-FDG PET images were resliced to match the MRI presentation. All ¹⁸F-FDG PET images were interpreted by a single observer with expertise in PET imaging of brain tumors. The interpreting physician was aware of patients’ clinical characteristics but was generally unaware of patients’ conditions over the course of radiotherapy. In accord with prior published studies (13–15), ¹⁸F-FDG PET images were interpreted by comparing uptake in the tumor region with average uptake in normal white matter, defined as the centrum semiovale contralateral to the tumor.

Areas meeting any of 3 criteria were considered to represent metabolically active tumor. The first criterion was the presence of greater uptake of ¹⁸F-FDG in tumor than in normal white matter. To facilitate this determination, images were displayed with both standard gray-scale windowing and with the setting of windowing thresholds to normal white matter uptake levels, as defined above.

The second criterion was the presence of abnormal ¹⁸F-FDG uptake within the boundaries of the abnormal region on T2-weighted MR images. Elevated ¹⁸F-FDG uptake in areas appearing normal on MRI was considered to be functional rather than tumor based. An example is increased uptake in the visual cortex when it appeared normal on MRI.

The third criterion, applied to the special case of tumor sites in gray matter, when both normal brain and metabolically active tumor can have uptake greater than that of white matter, was that only regions with uptake greater than that of white matter and also within the boundaries of MRI contrast enhancement were considered positive. Areas of ¹⁸F-FDG uptake in gray matter abnormal only on T2-weighted MRI cannot be reliably classified as tumor versus damaged or edematous gray matter and were therefore not included in the ¹⁸F-FDG-avid tumor volume. In contrast, areas with increased ¹⁸F-FDG uptake with contrast enhancement indicate tumor, as opposed to the alternate possibilities for a contrast-enhancing region: postsurgical changes or necrosis.

Contours were placed for the ¹⁸F-FDG PET-positive boost region by coregistering the ¹⁸F-FDG PET images to the treatment planning CT scan using the image manipulation software of Prism. Because patients were scanned in the same head immobilization mask for PET as for CT, only translation and not rotation was allowed in the coregistration process. All coregistrations were confirmed visually by a single experienced observer. Estimated coregistration accuracy, by a comparison of common visualized structures (eye musculature, outer cortical boundaries, etc.), was 1–2 pixels (2–4 mm).

¹⁸F-FDG PET tumor volume contours were drawn on the CT-coregistered ¹⁸F-FDG PET images using the Prism software and with reference to the original ¹⁸F-FDG PET images. The sites of ¹⁸F-FDG-avid tumor, identified as described above, were delineated by contours with the following considerations. Contours were drawn to closely encircle all sites of ¹⁸F-FDG-avid disease. For lesions with a cold center by ¹⁸F-FDG PET, contours were drawn to include the outer rim of increased uptake and to exclude areas in the center without increased uptake. Whenever possible, contours were made to be contiguous within a slice. However, for lesions in which separate, noncontiguous sites of ¹⁸F-FDG uptake were seen in the tumor bed, noncontiguous contours were drawn, and the ¹⁸F-FDG PET tumor volume was taken as the sum of the volumes of the noncontiguous contours. Once contours were drawn on the individual slices, they were checked for slice-to-slice consistency and adjusted as necessary. In addition, the ¹⁸F-FDG PET contours were checked for consistency with the T2-weighted MRI contours to make sure the ¹⁸F-FDG PET contours were within the T2-weighted MRI contours to meet the criteria for identifying areas of abnormal tumor uptake.

Final noncoplanar fields (3–4) were designed to treat the volume of abnormal ¹⁸F-FDG uptake plus a 0.5-cm margin (the third planning target volume). This final boost volume was treated for an additional 20 Gy (2 Gy per fraction), for a total dose of 79.4 Gy in 43 fractions. When critical structures such as the optic chiasm or optic nerve abutted the final boost volume, margins were adjusted to respect tolerance of these critical structures. If no abnormal ¹⁸F-FDG uptake was observed, the treatment was stopped after the conventional course of 59.4 Gy. After completing radiation therapy, patients were followed every 3 mo by clinical examination and MRI. If increased gadolinium enhancement was observed, patients were strongly encouraged to undergo a repeated ¹⁸F-FDG PET scan to verify disease progression. ¹⁸F-FDG PET confirmed progression if the areas of new contrast enhancement on MRI also had ¹⁸F-FDG uptake greater than that of white matter and either not seen on the planning study or qualitatively worse than on the planning study.

Analysis of Correlation Between PET and MRI Volumes

¹⁸F-FDG PET volumes were analyzed for size and location and were compared with the MRI-based volumes. Volumes were determined by summation of digitized contours from axial images in Prism. Volumes included in the analysis were T2-weighted MRI signal abnormality, T1-weighted MRI gadolinium enhancement, T1-weighted MRI gadolinium enhancement plus operative cavity, and ¹⁸F-FDG PET abnormal activity. Differences in volumes were initially assessed by size criteria using a simple ratio between volumes (i.e., PET volume to MRI volume). When absolute volumes were similar (<25% difference), volume locations were assessed and the degree of volume overlap was determined by a concordance index (CI) as CI = ([P + M]/X) − 1, in which P is the boost target volume determined by PET, M is the boost target volume determined by MRI, and X is the volume encompassing the combined ¹⁸F-FDG PET (P) and MRI (M) abnormalities. Complete concordance, or equivalence between volumes P, M, and X, would yield a CI of 1 (i.e., P + M/X = 2). Conversely, complete nonconcordance would yield a CI of 0 (i.e., volume P and volume M would be entirely separate, and thus P + M/X = 1). For example, if the MRI volume measured 10 cm³ and the PET volume also measured 10 cm³, by absolute volume the targets would be equivalent. However, to assess volume overlap (concordance), the CI would be applied. If the MRI volume and PET volume were indeed identical, the CI would result in a value of 1 [([10 + 10]/10) − 1 = 1]. Alternatively, if the volumes were entirely discordant (no overlap), the CI formula would yield a value of zero [([10 + 10]/20) − 1 = 0]. For all other scenarios, the CI would yield a value between 0 and 1, with higher values meaning greater concordance. Values of CI less than 0.75 indicate discordance exceeding 25%. A CI of 1 was assigned if both P and M equaled 0.

Analysis of Potential Prognostic Factors

Age, Karnofsky Scale, T2-weighted MRI volume, T1-weighted MRI gadolinium enhancement volume (both with and without inclusion of the resection cavity), and ¹⁸F-FDG PET volume were analyzed as prognostic variables for TTP and survival from date of diagnosis using the Cox proportional hazards model for continuous variables. Statistical analyses were performed using the StatView (1986 version; Abacus Concepts, Inc., Berkeley, CA) software package.