This is a page that curates AI-related papers published worldwide. All content here is summarized using Google Gemini and operated on a non-profit basis. Copyright for each paper belongs to the authors and their institutions; please make sure to credit the source when sharing.
Zacharia Mesbah, L eo Mottay, Romain Modzelewski, Pierre Decazes, S ebastien Hapdey, Su Ruan, S ebastien Thureau
Outline
The 2024 AutoPET competition aimed to develop a fully automated lesion segmentation algorithm for PET/CT scans using FDG or PSMA-based tracers, without knowing the tracer type. This paper describes how to train two sets of six-model ensembles using the nnUNetv2 framework and how to select the model set for segmentation using MIP-CNN.
Takeaways, Limitations
•
Takeaways: We present an effective solution to the problem of lesion segmentation in PET/CT scans using various tracers, leveraging the nnUNetv2 framework. A model selection strategy using MIP-CNN enhances adaptability to various tracers.
•
Limitations: This paper focuses on the description of a specific framework and model, and lacks comparative analysis with other approaches. Analysis of the impact of variables other than various tracers (e.g., scan quality, lesion size) is inadequate. Performance validation results in a real-world clinical setting are not presented.
