Advancing MRI segmentation with CLIP-driven semi-supervised learning and semantic alignment

Bo Sun, Kexuan Li, Jingjuan Liu, Zhen Sun, Xuehao Wang, Yuanbo He, Xin Zhao, Huadan Xue, Aimin Hao, Shuai Li, Yi Xiao

Abstract: Highlights•This study first integrate CLIP for semi-supervised medical image segmentation.•Proposed a novel CLIP-driven framework specialized for consistency learning.•Proposed a novel module to better align and represent text and images.•Designed a module for consistent, comprehensive segmentation across multi-axial MRIs.•Three multi-modal datasets were constructed to demonstrate the performance.