SAM-Geo3D: A Geometrical Method to Extend SAM to 3D

Jikai Zhang, Zafer Yildiz, Hanxue Gu, Haoyu Dong, Maciej A Mazurowski

Keywords: Segmentation, Segment Anything Model, 3D SAM

Abstract: Segment Anything Model (SAM) offers a promising approach for image segmentation tasks. However, SAM works in 2D making it less useful when segmenting cross-sectional images, such as MRIs. To address this, we proposed SAM-Geo3D, a geometrical method that extends SAM into the 3D manner. Given a few prompt points on a target component, SAM-Geo3D segments the component through all slices in 3D without onerous deep-learning-based training. We validated SAM-Geo3D on five knee MRI volumes. Results showed that SAM-Geo3D outperforms SAM when using the same, limited number of input prompt points.

Submission Number: 156