The autoPET challenge: Towards fully automated lesion segmentation in oncologic PET/CT imaging

Abstract

Abstract We describe the results of the autoPET challenge, a biomedical image analysis challenge aimed to motivate and focus research in the field of automated whole-body PET/CT image analysis. The challenge task was the automated segmentation of metabolically active tumor lesions on whole-body FDG-PET/CT. Challenge participants had access to one of the largest publicly available annotated PET/CT data sets for algorithm training. Over 350 teams from all continents registered for the autoPET challenge; the seven best-performing contributions were awarded at the MICCAI annual meeting 2022. Based on the challenge results we conclude that automated tumor lesion segmentation in PET/CT is feasible with high accuracy using state-of-the-art deep learning methods. We observed that algorithm performance in this task may primarily rely on the quality and quantity of input data and less on technical details of the underlying deep learning architecture. Future iterations of the autoPET challenge will focus on clinical translation.