PulseAugur
LIVE 14:08:56
tool · [1 source] ·
0
tool

AI models struggle with unseen PET/CT tracer combinations despite segmentation gains

The autoPET3 challenge, held in conjunction with MICCAI 2024, focused on automated lesion segmentation in whole-body PET/CT scans, specifically testing compositional generalization. The challenge utilized a large dataset comprising over 1,600 PET/CT studies from two major hospitals, including the largest publicly available annotated PSMA PET/CT dataset. Seventeen teams developed algorithms, primarily based on nnU-Net, with the top performer achieving a mean Dice Similarity Coefficient (DSC) of 0.66. The results indicated that while in-domain segmentation is nearing human agreement, generalization to unseen tracer-center combinations remains a significant challenge, with algorithm choice having less impact than data heterogeneity among top performers. AI

Summary written by gemini-2.5-flash-lite from 1 source. How we write summaries →

IMPACT Highlights limitations in compositional generalization for medical image segmentation models, indicating areas for future research and development.

RANK_REASON This is a research paper detailing a challenge and its results in automated medical image segmentation. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.CV →

COVERAGE [1]

  1. arXiv cs.CV TIER_1 · Jakob Dexl, Katharina Jeblick, Andreas Mittermeier, Balthasar Schachtner, Anna Theresa St\"uber, Johanna Topalis, Maximilian Rokuss, Fabian Isensee, Klaus H. Maier-Hein, Hamza Kalisch, Jens Kleesiek, Constantin M. Seibold, Hussain Alasmawi, Lap Yan Lennon ·

    The autoPET3 Challenge -- Automated Lesion Segmentation in Whole-Body PET/CT - Multitracer Multicenter Generalization

    arXiv:2605.05775v1 Announce Type: new Abstract: We report the design and results of the third autoPET challenge (MICCAI 2024), which benchmarked automated lesion segmentation in whole-body PET/CT under a compositional generalization setting. Training data comprised 1,014 [18F]-FD…