artificial intelligence, image fusion, image registration, learning (artificial intelligence), Electrical engineering. Electronics. Nuclear engineering, TK1-9971

Abstract Medical image registration is a fundamental and important technique in the field of medical image analysis. This study proposes a novel unsupervised end‐to‐end registration network, aiming to enable the model to actively acquire image features in the field of medical imaging with limited samples, which efficiently integrates multi‐scale features to achieve higher accuracy in registration. By utilizing region‐to‐region routing, this model actively preserves the most relevant features of the images, thereby improving training and learning efficiency. The model is evaluated by several publicly available datasets. The new network proposed in this study achieved the best registration accuracy among various advanced traditional and learning‐based methods.