Advanced medical image processing, encompassing critical tasks such as 3D image registration and precise lesion segmentation, is a fundamental component of modern clinical workflows. It plays a pivotal role in surgical navigation, tumor tracking, and multi-modal pathology analysis. Despite significant advancements in Artificial Intelligence, developing algorithms that are simultaneously highly accurate, topologically robust, and computationally efficient remains a formidable challenge. This Special Issue seeks to address these critical bottlenecks by exploring cutting-edge deep learning architectures. We invite original research and comprehensive reviews focusing on innovative paradigms, including wavelet encoding, deformable feature interaction, Transformer and Mamba architectures, and multi-scale feature fusion. The primary goal is to highlight breakthrough methodologies that enhance medical image segmentation and diffeomorphic registration, while maintaining lightweight structures suitable for real-time clinical deployment. By bridging the gap between complex algorithmic design—such as multi-modal image synthesis and edge-aware networks—and practical clinical applicability, this Special Issue aims to advance the frontiers of AI in medicine and oncology, ultimately facilitating more precise diagnostic and therapeutic interventions.