Structurally Consistent MRI Colorization using Cross-Modal Fusion Learning

Medical image colorization enhances the interpretability of imaging data by providing clearer anatomical insights. We aim to transfer diverse anatomical colors from Cryosection data to MRI scans while preserving MRI structural details. Our novel architecture achieves structurally consistent color transfer by fusing Cryosection segmentation semantics into the colorization process. This enables stable organ-level colorization in MRI images without requiring precise registration between the input MRI and Cryosection data or segmentation of the MRI itself. Additionally, we introduce a feature compression-and-activation mechanism that captures global organ-level information, suppresses noise, and facilitates accurate and realistic organ-specific colorization of MRI scans. Comparative experiments with state-of-the-art techniques demonstrate significant visual improvements, confirming the effectiveness of our method.