Abstract

Info

Author(s):: YANG Zehua1; SUN Zheng1; 2; 1. Department of Electronic and Communication Engineering, North China Electric Power University, Baoding 071003, China 2. Hebei Key Laboratory of Power Internet of Things Technology, North China Electric Power University, Baoding 071003, China

Abstract:: Abstract: Multimodal medical image fusion integrates complementary information derived from CT, MRI, PET, and other modalities to support clinical diagnosis and treatment planning. However, discrepancies in imaging mechanism, resolution, and noise across different modalities make fusion and evaluation challenging. Herein a systematic review of traditional and deep learning-based fusion methods is provided. It systematically introduces classical approaches such as multi-scale transformation, sparse representation, and component decomposition, along with an analysis of their inherent limitations, and then summarizes the core strategies and prominent advantages of deep learning frameworks including CNN, GAN, autoencoder, and Transformer. Furthermore, it elaborates on the pixel-level, structural/information feature-level, and task-driven evaluation metric systems, and reviews their clinical applications in oncology and neurological diseases. Finally, major challenges are discussed, including data scarcity, registration sensitivity, lack of standardized evaluation criteria, and insufficient interpretability. The prospects of self-supervised pretraining and standardized clinical validation are also offered, aiming to providing a reference for technological innovation and clinical translation.