Automated PET-MR segmentation models for glioblastoma reveal complementary FET-PET and MR information for recurrence

Automatic nnU-Net segmentation of glioblastoma PET/MR achieved Dice scores of 0.76 (MR-enhancing), 0.69 (edema), 0.71 (FET-PET uptake). MR and FET-PET recurrence delineations differed significantly (DSC 0.45, p=0.0497). MR radiomic features could not spatially identify PET findi…

• Model performance on test set: DSC 0.76±0.24 (MR-enhancing), 0.69±0.23 (MR-edema), 0.71±0.20 (PET-uptake), 0.93±0.05 (planning target volume), 0.70±0.13 (organs-at-risk).
• Manual MR and FET-PET recurrence delineations differed in size (p=0.0497) and location (DSC 0.45±0.20).
• None of the 37 MR radiomic features with significant MR∩PET vs MR-Only differences allowed spatial identification of PET findings (positive predictive value and sensitivity <0.6).

Automated summary

RadPigeon summaries are original and for information only. They are not clinical advice.