Problem
Research questions and friction points this paper is trying to address.
Addresses pathological image registration challenges
Focuses on focal tissue impact in registration
Proposes unsupervised GIRNet for accurate registration
Co-learning Semantic-Aware Unsupervised Segmentation for Pathological Image Registration
🤖 AI Summary
Pathological image registration suffers from disrupted spatial correspondence and deformation distortion caused by lesions, compounded by the absence of pixel-level annotations and anatomical consistency constraints. To address these challenges, we propose the first unsupervised co-learning framework explicitly designed for lesion robustness. Our method establishes bidirectional semantic constraints between registration and semantic segmentation tasks to jointly model anatomical consistency and align domain-invariant features. It employs a deformable convolutional U-Net for registration, a contrastive learning–driven segmentation branch, and a composite loss integrating mutual information maximization and deformation regularization. Evaluated on the BraTS and PANDA datasets, our approach achieves a 12.6% improvement in registration accuracy and a Dice score of 0.84—substantially outperforming state-of-the-art unsupervised methods. This work advances pathological image registration by unifying structural fidelity, lesion-invariant feature learning, and geometric regularization within a single, annotation-free framework.
Innovation
Methods, ideas, or system contributions that make the work stand out.
Unsupervised learning for pathological image registration
Co-learning integrates segmentation and inpainting
GIRNet framework improves lesion identification accuracy
🔎 Similar Papers
No similar papers found.
