Medical image fusion based on improved multi-scale morphology gradient-weighted local energy and visual saliency map

Medical image fusion refers to the process of fusing images of different modes of the same object using image processing technology to maximize the mining of image information and improve image quality. At present, many methods are available to study image fusion, but they usually have shortcomings such as low image contrast and a weak ability to retain image details and edges. To solve these problems, we propose a new multimodal medical image fusion method. In this algorithm, the original image is decomposed into high frequency and low-frequency information by non-subsampled shearlet transform (NSST). Low-frequency information is fused by visual saliency maps, which avoids edge loss caused by direct use of the coefficient maximum fusion rule. High-frequency information is fused by a method jointly guided by the improved multi-scale morphology gradient and weighted sum of eight-neighborhood-based modified Laplacian, which retains the texture details and the edge of the image. Finally, the fused image is generated by the NSST inverse transform. This strategy solves the problem of insufficient detail extraction in traditional algorithms, improves the overall appearance of the fused image, and enhances the contrast. To verify the effectiveness of the algorithm, we applied this technique to four different medical image modality combinations, compared the results with nine image fusion methods published in recent years, and evaluated the fused images using image quality evaluation indexes. Our algorithm achieved better results in terms of subjective vision and objective image quality evaluations and therefore should be competitive with existing technologies.