Abstract:The previously proposed medical image fusion algorithms extracted the same level of features from the source image, ignoring the unique features of the source image. To solve the problem, this work proposed a fusion method to extract unique features from different modal medical images. Firstly, the improved multi-level decomposition based latent low-rank representation method was used to extract the basic information and detailed information of CT and MRI images, and further extracted the bone contour information of CT image and soft tissue detail information of MRI image according to the different imaging principles. Then, this work proposed a local information entropy-weighted local energy function method to fuse the detail information, and utilized the structural saliency and sum of eight-neighborhood based modified laplacian to fuse the basic information. Finally, an image-guided enhancement method was proposed to enhance the fused base layer and detail layer with unique features as the guide. Experiments showed that, compared with the representative fusion methods of recent years, this approach not only improves the objective evaluation indicators of AG, EPI, VIF and SD by 9.45%, 11.75%, 14.79% and 10.51%, respectively, but also achieves better results in subjective evaluation, and realizes the accurate fusion of CT and MRI images.