Deep Learning–Based Image Registration and Normalization for Robust Multi-modal Medical Image Fusion

The integration of CT, MRI, and PET via multimodal medical image fusion is establishing itself as an effective tool to enhance diagnostic accuracy by leveraging complementary information from different imaging modalities. Yet, effective image fusion relies on image registration and consistent intensity normalization, which remain challenging because these factors depend on differences in spatial resolution, contrast, and acquisition protocols across imaging modalities. In this paper, we explore a deep learning–based method as a potential solution for robustly registering and normalizing multimodal medical images, enabling reliable image fusion. In the proposed method, a convolutional neural network is used to align and standardize the intensities of multi-modal images before fusion, ensuring consistency in spatial distribution and intensity. Fusion and registration should be further complemented and augmented by correcting cross-modal spatial alignments and modulating representations to address intermodal misalignment and intra-modal incoherence. We detail the positive impact our proposed approach, particularly in addressing fusion and registration, has on rapid diagnosis and downstream diagnostic performance. We detail the positive impact our proposed approach, particularly in addressing fusion and registration, has on rapid diagnosis and downstream diagnostic performance