Research On Medical Image Fusion Algorithms Based On Multi-scale Analysis

Medical image fusion technology integrates multi-modal medical image feature information to get more detailed and comprehensive description,providing important visual evidence for medical research and clinical diagnosis.It is a combination of medical imaging and information fusion technology.Aiming at the application of CT and MRI,MRI and PET image fusion in medical images,the theory of multi-scale analysis combined with image processing methods such as minimum gradient smoothing filter,non-subsampled contourlet transform,structure tensor,nonlinear HIS transform and pulse coupled neural networks,are studied for the fusion of medical images.The main research contents are summerized as follows:Based on the advantages of structure tensor in describing geometric features of images and non-subsampled contourlet transform in describing multi-scale and multi-directional features,a medical image fusion algorithm based on structure tensor and non-subsampled contourlet transform is proposed.Firstly,the structure tensor of the source image is pre-fused according to the local saliency feature measure.The low frequency subband and high frequency subband of non-subsampled contourlet transform are pre-fused according to the weighted average method and the maximum improved sum of modified Laplacian method respectively.Then,the optimization model is constructed with the constraints of the gradient of the fused image and the coefficients of non-subsampled contourlet transform close to the pre-fused gradient and coefficients.The conjugate gradient algorithm is applied to solve the NP-hard problem to obtain the fused image.Based on the edge preserving property of gradient minimum smoothing filter and the sparse representation of smooth images,a novel algorithm combining multi-scale gradient minimum smoothing filter decomposition and sparse representation is proposed for medical image fusion.First,the source image is decomposed into the base image and the detail image by the multi-scale gradient minimum smooth filter framework.An over-complete dictionary and two sparse coding matrices are obtained by joint sparse representation of the base images.In order to get the fused base image,the energy of the coefficient energy and the column energy in the sparse coding coefficient fusion is taken into consideration,and the gradient in the spatial domain is simulated to make the comprehensive decision.In order to get the fused image,the energy of the detail image is extracted and the energy of the edge of the region is weighted,and the decision is based on the activity.For the fusion application of gray-level MRI image and pseudo-color PET image,a novel image fusion algorithm based on nonlinear HIS transform,non-subsampled contourlet transform and improved pulse coupled neural network is proposed.The PET image is converted from RGB to HIS color space by taking advantage of the color fidelity of HIS transform,and the intensity component is extracted to be fused with the MRI image in the gray space.Then,the images are decomposed into low frequency subbands and high frequency subbands through non-subsampled contourlet transform.For low frequency subband images,the local Gaussian membership in fuzzy mathematics is applied for fusion.For high frequency subband images,the improved pulse coupled neural network is applied for fusion.The fused image features remain more complete and the details are more abundant.These algorithms can be applied to medical image fusion and more effectively assist clinical diagnosis.