Medical Image Alignment for Different Resolutions and Fields of View Using Contrastive Learning with Feature-Level...

This study proposes a feature-space-based alignment method that enables effective alignment across images of different scales and modalities. Alignment refers to the process of calculating the similarity between images and accurately superimposing them based on this similarity. The proposed method addresses challenges in conventional approaches by achieving modality-independent and noise-resilient similarity calculation. Additionally, we incorporate contrastive learning to enhance the feature representation and improve the alignment process by learning discriminative features. Experiments using datasets simulating real-world conditions demonstrate the robustness of the proposed method against scale differences and noise.