Segment Anything Model for Medical Image Analysis: an Experimental Study

https://github.com/mazurowski-lab/segment-anything-medical

[Submitted on 20 Apr 2023 (v1), last revised 25 Apr 2023 (this version, v2)] on arkiv

Dataset

Experiments conducted mostly on MRI, X-ray and CT datasets.

Results

Methods of using SAM

1. Semi-automated annotation
2. Assisting other Segmentation Models
3. New Medical Image Foundation Segmentation Models

Segment Anything in Medical Images

https://github.com/bowang-lab/MedSAM

[Submitted on 24 Apr 2023] on arkiv

다양한 모드들이 존재.

Segment everything

bounding box

point mode: 1,2,3

Medical image 특성상 bounding box모드를 기준으로 MedSAM train

MedSAM

Endoscope dataset에 대하여 좋은 segmentation 성능을 보임