Using the SAM2 Foundation Model for Zero-shot Delineation of MRI-detected Brain Metastases for Stereotactic Radiosurgery

Motivation: Due to their intricate morphology, brain metastases (BM) present significant challenges in detection and delineation during radiotherapy treatment planning, impacting the radiosurgery quality. Goal(s): This study explores zero-shot applications of Segment Anything Model 2 (SAM2) to delineate BM on contrast-enhanced T1-weighted MRI, aiming to enhance the accuracy and efficiency of treatment planning. Approach: SAM2, a transformer-based foundation model, was applied to MRI from 56 patients, utilizing its memory attention mechanism for tracking BM across imaging sessions. Results: SAM2 performed well in delineating larger BM, achieving a Dice coefficient of 0.772±0.215. However, it faced the challenges in segmenting multiple smaller lesions under 0.12 cm3. Impact: This brain metastases delineation tool has been shown to enhance the efficiency of treatment planning, potentially improving the effectiveness of stereotactic radiosurgery for brain metastases. Such advancements could lead to better clinical efficiency and patient outcomes.