Fusion Of PET And MRI Images Based On Improved Structure Tensor

PET images of patients with Alzheimer's Disease can reflect the metabolic changes in the brain,however,the resolution is too low to see the structure of the diseased part.The high resolution of MRI images can show clear tissue structure,but it does not show the metabolism of the tissue structure.Then fuse PET and MRI images of Alzheimer's Disease patients,so as to determine the specific site of metabolic abnormalities and improve the diagnostic accuracy.Selection of the image fusion methods based on structural tensor based on the spatial characteristics of 3D images based on PET and MRI.This fusion method is simple and efficient and the fused image retains many edge and detail information.At present,the problems in 3D image fusion based on structural tensor are: the calculation of the gradient does not fully consider the influence of the neighborhood voxels,resulting in low gradient accuracy,the randomness of the threshold setting during image decomposition makes the fusion result unstable,the selection of the fusion rule has a direct impact on the quality of the fused image.The direct algorithm for the gradient of the structure tensor considers the problem of fewer voxels in the neighborhood.Therefore,the two-dimensional Sobel operator is introduced into the gradient solution of three-dimensional PET and MRI images.In order to better adapt to the spatial characteristics of 3D images,it is extended to 3D Sobel operators.Random selection of thresholds in image decomposition based on structural tensor makes the decomposition result random.In order to solve this problem,we propose the best decomposition based on traces.The optimal threshold of the trace is determined according to the change trend of the threshold of the trace and the fitted curve of the evaluation index and the visual effects of the decomposed and fused images.The fusion rule based on the Weighted Sum of Eight-neighborhood-based Modified Laplacian has a low calculation accuracy when calculating the energy of 3D images.Therefore,the 8-direction energy of gradient is used to replace the Weighted Sum of Eight-neighborhood-based Modified Laplacian to improve the performance of the fused image.Fusion of PET and MRI medical images of patients with Alzheimer's disease.The experimental results show that the fusion effect of the image fusion algorithm based on the structure tensor composed of the gradient calculated by the three-dimensional Sobel operator is better than that of the direct gradient calculation method.The image obtained by using the best trace-based decomposition is more stable than a decomposed image obtained by randomly setting a threshold.Energy of gradient as an energy input term of Weighted Sum of Eight-neighborhood-based Modi fied Laplacian fusion rule is superior in subjective and objective evaluation.