Research On Multi-scale Fusion Of Medical Imaging Data At Pixel-level

Fusion of medical imaging data is the process of merging multiple images from a single or multiple imaging modalities to improve the imaging quality with preserving the specific features.Fusion of medical imaging data is to increase the clinical applicability of images for diagnosis and assessment of medical problems.We focus on multi-scale medical image fusion methods at pixel-level.Usually,multi-scale medical image fusion methods at pixel-level include three parts: image decomposition-reconstruction,image fusion rule,and image quality metric.To solve the problems of the existing fusion methods in terms of image decomposition-reconstruction and image fusion rule,four new multi-scale medical image fusion methods at pixel-level are proposed in this thesis.Two-scale intrinsic image decomposition based MRI-PET method is proposed for solving the problems of high computational time and image noise using the existing image fusion methods.The contribution of the proposed two-scale method is to construct a fast method while introducing little noise in the spatial domain.The proposed method firstly adopts Retinex theory for MRI to get its two-scale intrinsic image representation,and adopts Grey-World theory for PET to get its two-scale intrinsic image representation.Then,principle component analysis,importance of image coefficients,and intensity-hue-saturation are used to fuse the decomposed images,for respectively.Experimental results show that the proposed two-scale intrinsic image decomposition based method could recover reflectance image from the input image while reducing image noise.Three-scale structure tensor based MRI-PET and MRI-SPECT method is proposed for solving the problem of intensity loss and color distortion.In the proposed method,two different structure tensors are used as the image decomposition-reconstruction tool.The proposed method firstly adopts structure tensor to decompose MRI into its three-scale image representation,and adopts color structure tensor to decompose PET/SPECT into its three-scale image representation.Then,absolute maximum is used to fuse smooth images.And spatial frequency is used to fuse intensity images and detail images.Experimental results show that three-scale structure tensor based method could preserve the intensity from both MRI and PET/SPECT.Multi-direction feature based MRI-CT,MRI-PET,and MRI-SPECT method is proposed for solving the problem of edge artifacts in the fused images using Laplacian pyramid transform.In the proposed method,multi-redirection feature is applied as the image fusion rule of the decomposed residual images by Laplacian pyramid transform in the frequency domain.Firstly,the input image is decomposed into residual images and base images by Laplacian pyramid transform.In regard to image fusion rule,affine transform is used to extract multi-direction information from residual images.And then,the residual images with multi-direction are enhanced by outline and contrast features.Experimental results show that the fused image by multi-direction based method is in high contrast and clear edge while without block artifacts.Multi-saliency feature based MRI-CT and MRI-PET method is proposed for solving the problem of edge artifacts and color distortion resulting from local extreme scheme.In the proposed method,multi-saliency feature is used as the image fusion rule of the decomposed smooth images and the decomposed detail images.Firstly,the input image is decomposed into smooth images and detail images at different scales by local extrema scheme.In regard to image fusion rule,edge saliency feature is use to fuse the smooth images of MRI/CT.And,color saliency feature is used to fuse the detail images of PET.Experimental results show that edge saliency feature works well at removing edge artifacts and color saliency feature works well at removing color distortion.