Open-Source Deep Learning Pipeline Automates PET-Only Tumor Segmentation

Researchers developed LION, an open-source PET-only segmentation pipeline for [F]FDG and PSMA-targeted PET/CT, trained and validated on over 7,000 scans across multiple centers. The study reports that incorporating organ context improved median Dice scores (0.62 to 0.71 for FDG; 0.75 to 0.83 for PSMA), that spatial diversity of training data (measured via tumor occurrence maps) correlated strongly with segmentation performance, and that LION outperformed other open-source tools on a multicenter holdout cohort. The authors state that a mixed-disease model trained on fewer cases matched a disease-specific model trained on substantially more data.

Most relevant for: Relevant to oncologic imaging workflows using FDG-PET or PSMA-targeted PET/CT for tumor segmentation across multiple cancer types, including prostate cancer and lymphoma.

Summary written by an AI model from the source text and dated by the source's publication date. Check the original before acting on it. How MedFollow selects and summarizes updates.

Informational only. Not medical advice. Not a substitute for clinical judgment. Always consult current guidelines and your own clinical judgment.