Research On Color Perception Of Medical Images Based On Hierarchical Manifold Learning

The color perception of medical images is a software technology,which perceives high-dimensional medical data as a color-coded image,thus enhancing the image representation and improving the visual effect of the medical worker.The traditional color perception methods using manifold learning or color transfer algorithms suffer from high computational cost,vague regional structure and lack of image another features.In order to quickly realize the color perception of medical images and fully express the image information,this thesis proposes the novel methods for color perception of diffusion tensor images and medical images by introducing manifold learning and algebraic multigrid.Meanwhile,the theory and experimental results demonstrate their effectiveness.Specific work as follows:1.Color perception of diffusion tensor images(DTI)based on manifold learning relates to the shortest path calculation and eigenvalue decomposition for all pixels,which renders complexity high.To address the issue,a novel approach for color perception of DTI using hierarchical manifold learning is proposed.The approach introduces algebraic multigrid to establish hierarchical structure of pixels.Only a few nodes on the top layer are dimension reduction and mapped to the color space,so that the high dimension DTI data is presented by a color image.What’s more,the proposed algorithm involves a new distance metric to estimate the pixels on the top layer,guaranteeing accurate distance and avoiding repeating calculation.The algorithm is applied to the synthetic DTI and real DTI,and theory analysis demonstrates that the method can reduce the computation complexity fromO(N~3)(color perception of DTI based on ISOMAP)toO(N~2)and the experimental results have lower information entropy,which means image distribution is more ordered.2.Most of the existing medical images are presented by gray images,which are single feature,thus leading to the image information can not be expressed sufficiently.A novel approach for color perception of medical images based on multiple features fusion is proposed.The fusion algorithm extracts gradient feature in multiple direction and fuses brightness feature to retain more image information;the algorithm establishes pyramid structure to reduce the amount of calculation by selecting representative pixels in similar region.The low-dimensional coordinates are calculated using manifold-based techniques and interpolation from top to down.Last,the color perception of medical images is obtained by projecting the low-dimensional coordinates to RGB color space.The method is applied to the normal MRI,CT images and MRI with disease.The experimental results demonstrate that the fusion images have rich color information,and better definition and contrast between target and background,comparing with some color transfer methods.3.Designing and implementing a simulation system for medical images based on color perception.In this system,the two proposed methods and the contrast methods are integrated to realize the interface visualization.Users can select the object images manually(diffusion tensor images/conventional medical images).Then according to different requirement,choosing the corresponding color perception method,and displaying the results,in order to analyze proposed approaches conveniently.