Segment Anything Model (SAM) for Radiation Oncology

In this study, we evaluate the performance of the Segment Anything Model (SAM) model in clinical radiotherapy. We collected real clinical cases from four regions at the Mayo Clinic: prostate, lung, gastrointestinal, and head \& neck, which are typical treatment sites in radiation oncology. For each case, we selected the OARs of concern in radiotherapy planning and compared the Dice and Jaccard outcomes between clinical manual delineation, automatic segmentation using SAM's "segment anything" mode, and automatic segmentation using SAM with box prompt. Our results indicate that SAM performs better in automatic segmentation for the prostate and lung regions, while its performance in the gastrointestinal and head \& neck regions was relatively inferior. When considering the size of the organ and the clarity of its boundary, SAM displays better performance for larger organs with clear boundaries, such as the lung and liver, and worse for smaller organs with unclear boundaries, like the parotid and cochlea. These findings align with the generally accepted variations in difficulty level associated with manual delineation of different organs at different sites in clinical radiotherapy. Given that SAM, a single trained model, could handle the delineation of OARs in four regions, these results also demonstrate SAM's robust generalization capabilities in automatic segmentation for radiotherapy, i.e., achieving delineation of different radiotherapy OARs using a generic automatic segmentation model. SAM's generalization capabilities across different regions make it technically feasible to develop a generic model for automatic segmentation in radiotherapy.