CT and MRI Image Fusion Based on Improved Denoising Diffusion Probability Model

Deep learning methods have strong feature extraction capabilities in the field of multimodal medical image fusion. However, the current popular multimodal image fusion methods mostly take the registered medical images as the research object, while the medical images in real life do not meet this condition, so the fusion methods have the problems of complex process, time-consuming and lack of practicability. In order to solve the above problems, this paper studies a CT and MRI image fusion algorithm based on an improved denoising diffusion probability model. The algorithm uses the improved Demons method to perform non-rigid registration of CT and MRI images, and then adopts the optimized denoising diffusion probabilistic model (DDPM) and pulse coupled neural network (PCNN) hybrid model to further improve the ability to capture image features. The registration and fusion of CT images and MRI images are realized in an integrated manner. The results show that the algorithm has high registration accuracy, the structural similarity index of image fusion is increased by 48.4% on average, AG by 20.2% on average, MI by 17% on average, and peak signal-to-noise ratio by 3.6%, which realizes more efficient medical image registration and fusion, and is more practical.