Table 1.

Comparison of SAM approaches regarding applicability for medical imaging, suitability for 3D volumetric data, and computational efficiency of the core components: image encoder, prompt encoder, and mask decoder.

Method	Medical	Volumetric3D Data	Efficient Image Encoder	Efficient Prompt Encoder	Efficient Mask Decoder
SAM [16]	✘	✘	✘	✘	✘
MobileSAM [32]	✘	✘	✔	✘	✘
TinySAM [24]	✘	✘	✔	✔	✔
MedSAM [18]	✔	✘	✘	✘	✘
SAM-Med2D [4]	✔	✘	✘	✘	✘
SAM-Med3D [27]	✔	✔	✘	✘	✘
FastSAM3D (ours)	✔	✔	✔	✔	✔

The vanilla SAM [16] lacks in all criteria. MobileSAM [32] improves encoder efficiency, while TinySAM [24] accelerates all components, but neither addresses 3D medical imaging data. MedSAM [18] and SAM-Med2D [4] are tailored for medical 2D data yet do not improve efficiency. SAM-Med3D [27] handles 3D medical data but inference times for these limit or altogether preclude real-time interactive use with standard GPU hardware. Our proposed FastSAM3D meets all criteria, providing a comprehensive solution for efficient interactive medical image segmentation in volumetric 3D data.