A new contrast based multimodal medical image fusion framework

Abstract Medical image fusion plays an important role in clinical applications such as image-guided surgery, image-guided radiotherapy, noninvasive diagnosis, and treatment planning. The main motivation is to fuse different multimodal information into a single output. In this instance, we propose a novel framework for spatially registered multimodal medical image fusion, which is primarily based on the non-subsampled contourlet transform (NSCT). The proposed method enables the decomposition of source medical images into low- and high-frequency bands in NSCT domain. Different fusion rules are then applied to the varied frequency bands of the transformed images. Fusion coefficients are achieved by the following fusion rule: low-frequency components are fused using an activity measure based on the normalized Shannon entropy, which essentially selects low-frequency components from the focused regions with high degree of clearness. In contrast, high-frequency components are fused using the directive contrast, which essentially collects all the informative textures from the source. Integrating these fusion rules, more spatial feature and functional information can be preserved and transferred into the fused images. The performance of the proposed framework is illustrated using four groups of human brain and two clinical bone images from different sources as our experimental subjects. The experimental results and comparison with other methods show the superior performance of the framework in both subjective and objective assessment criteria.