AutoPET3 Challenge Benchmarks Lesion Segmentation in PET/CT

The third autoPET challenge, held at MICCAI 2024, evaluated automated lesion segmentation in whole-body PET/CT under compositional generalization. Training data included 1,014 [18F]-FDG studies from University Hospital Tübingen and 597 [18F]/[68Ga]-PSMA studies from LMU University Hospital Munich, forming the largest publicly annotated PSMA PET/CT dataset. The test set of 200 studies covered four tracer-center combinations, with two unseen pairings. A data-centric award category isolated data handling contributions using a fixed baseline model. Seventeen teams submitted 27 algorithms, mostly nnU-Net-based 3D networks with PET/CT channel concatenation. The top algorithm achieved mean DSC 0.66, FNV 3.18 mL, and FPV 2.78 mL.