IMG-76. FET PET reveals considerable volumetric and spatial differences in tumor burden compared to conventional MRI in recurrent glioblastoma

Abstract BACKGROUND In recurrent glioblastomas, changes in areas of contrast enhancement and the T2/fluid-attenuated inversion recovery (FLAIR) signal on conventional MRI represent the mainstay for local therapy planning. Nevertheless, compared to conventional MRI, the information on the tumor burden obtained from amino acid PET may be considerably different in terms of volumetric assessment and spatial orientation. METHODS At suspected recurrence, 56 patients with histomolecularly characterized glioblastoma underwent O-(2-[18F]-fluoroethyl)-L-tyrosine (FET) PET and MR imaging including contrast-enhanced and FLAIR sequences. Contrast-enhancing and FLAIR volumes were automatically segmented using HD-GLIO, and FET PET tumor volumes were assessed using the nnUNet-based JuST_BrainPET segmentation tool based on a tumor-to-brain ratio of ≥1.6. All segmentations were visually checked. Subsequently, an in-house developed workflow was used for a fully automated assessment of maximum and mean tumor-to-brain ratios. To evaluate spatial differences between the modalities, percentage overlap, the Dice similarity coefficient (DSC), and the 95th-percentile Hausdorff distance (HD95) were calculated. Recurrent disease was confirmed either by neuropathological evaluation of tissue obtained from surgery or stereotactic biopsy or prompted a change in treatment. RESULTS All patients had measurable disease according to the PET RANO 1.0 criteria (mean tumor-to-brain ratio, 2.2±0.2). In 52 patients (93%), the FET PET tumor volume was significantly larger than the contrast-enhancing volume (36.5±31.6 mL vs. 18.5±19.7 mL; P<0.001). On average, FET PET tumor volumes extended by 30% beyond the combined contrast-enhancing and FLAIR volumes. The spatial similarity between FET uptake and contrast enhancement was limited (mean DSC, 0.40±0.23), with an HD95 of 17.8±12.2 mm. The comparison of FET uptake with the FLAIR hyperintensity revealed even lower spatial similarity (mean DSC, 0.35±0.16), and a higher boundary discrepancy (HD95, 30.0±14.2 mm). CONCLUSIONS Our results strongly support integrating both imaging modalities into treatment planning of patients with glioblastoma at recurrence.