CT and MRI Fusion with Anisotropic Guided Filtering

The combination of CT and MRI can provide more accurate images of lesions, yielding a significantly higher diagnostic value compared to single-modality pathological images. However, in CT-MRI fusion, preserving the gray-scale distribution of the source image while avoiding 'detail halos' poses a challenge. Therefore, we propose the utilization of anisotropic guided filtering (AnisGF), which exhibits excellent edge-preservation properties, to address structural inconsistencies in regions between the two modalities. The local neighborhood variance is utilized for optimizing the weight to achieve maximum diffusion, and subsequently decomposing the source image based on this criterion. A pre-trained convolutional neural network (CNN) is employed to accomplish the mapping from the source image to the weight map, while AnisGF is utilized for maintaining local consistency between them. The efficacy of this novel image fusion algorithm in preserving intricate details without compromising has been demonstrated through a combination of qualitative and quantitative experiments.