A Multimodal Registration and Fusion Diagnostic System Based on Multi-scale Feature

The rapid growth in manufacturing and multimodal medical imaging industries, driven by evolving user needs, has led to an increasing demand for advanced digital medical imaging technology and corresponding diagnostic analysis systems. Designing human-centered algorithm systems, including digital consultation, intelligent medical care, and medical imaging for disease diagnosis and treatment tracking, becomes imperative. Given the variations in acquisition equipment and imaging technologies across different modal medical images, disparities in initial pose and spatial characteristics arise. To address this, a multimodal medical image registration algorithm is employed to harmonize data coordinates into a unified spatial coordinate system. This paper proposes a feature analysis and key point extraction network based on deep learning. Methods such as matrix decomposition and unsupervised iterative registration are utilized to achieve precise multimodal medical image registration. Leveraging this registration framework, the study focuses on analyzing latent mandibular deviation as a target disease. Features from diverse modal medical images are extracted and fused through a multi-level feature structure network, facilitating disease classification diagnosis and subsequent parameter analysis.