Meta released SAM 2, which can provide real-time, promptable object segmentation for static images and dynamic video content, integrating image and video segmentation functions. Medical image segmentation provides a basis for clinical diagnosis, etc. Segmentation based on neural network models is the mainstream method, but there are still challenges, such as model generalization and data difference problems. The Oxford University team developed a medical image segmentation model MedSAM-2 based on the SAM 2 framework, which treats medical images as videos and has a single prompt segmentation capability. The relevant paper has been published on the preprint platform arXiv. The paper comprehensively evaluates the model through the classification of different data sets, introduces the model architecture for effective segmentation processing of medical images of different dimensions, and the experimental results show that MedSAM-2 is fully ahead in performance and generalization ability. SAM helps medical image segmentation research to be hot, and many laboratories and academic teams are exploring its potential. For example, the team of Professor Ni Dong of Shenzhen University built the ultra-large-scale medical image segmentation dataset COSMOS 1050K and evaluated SAM. The teams of Fudan University and Shanghai Jiaotong University in Shanghai also studied SAM in the field of medical image segmentation.
Paper address:
https://arxiv.org/pdf/2408.0087
SA-V video segmentation dataset direct download:
https://go.hyper.ai/e1Tth
Medical SAM 2 sample medical segmentation dataset:
https://go.hyper.ai/TZZBj