The Decomposition And Registration For Breast Dynamic Contrast-Enhanced MRI

Breast cancer is a frequently-occurring disease of women. Early detection and treatment of breast disease are important for reducing mortality. Dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) has been an important tool for breast cancer detecting because of the high resolution for soft tissue. DCE-MRI produces a large number of images which provide abundant four-dimensional enhancement information of lesion. However, these images seriously burden physicians. The computer-aided diagnosis (CAD) system based on image processing and pattern recognition can assist radiologists in disease diagnosis. This thesis aims to improve the accuracy of CAD system by image decomposition and registration algorithm. To achieve this objective, a decomposition algorithm combined with total variation (DC-TV) is proposed. According to the different diffusion of contrast agent in normal and abnormal tissues, the DC-TV algorithm splits the mixed DCE-MRI into two different enhanced parts, which includes enhancement images of pure abnormal tissue and normal tissue. Thus, the algorithm we proposed provides an accurate diagnosis of lesion and a solution to non-rigid registration of DCE-MRI with intensities variation. The main work of this thesis is as followed:(1) The decomposed algorithm based on DCE-MRI. Based on the difference of contrast agent diffusing in different cells, the model of spatial continuity for abnormal cells and temporal continuity for normal tissues can be built. The DC-TV method we proposed innovatively decomposes original DCE-MRI into structure and texture images based on the above continuities, which distinguishes from the existing algorithm. Firstly, the quantitative evaluation of displacement is calculated by the correction of biased field and the dynamic threshold segmentation. Secondly, a total of 95 lesions without obvious displacement are decomposed to acquire pure enhancement images of abnormal tissue. Finally, the decomposition results are analyzed by various evaluation principles. Experimental results show that the texture images with temporal continuity can be achieved by DC-TV method. The time intensity curve (TIC) of structure image is more conform to Three-Time-Point (3TP) principle. Compared with the classification results utilizing original images, the ones using structure images show more accurate outcome of classifying benign and malignant tumor.(2) Registration based on breast DCE-MRI:The correction of the breast DCE-MRI with obvious displacement. The registration of breast DCE-MRI is one of the non-rigid registration with intensities variation, which is difficult to overcome. Inspired by the DC-TV, a novel registration method based on decomposition algorithm is proposed. By separating the normal and abnormal tissue from original DCE-MRI, the texture images with consistent intensity are acquired, which can be used to calculate the accurate optical field. The calculated deformation field based on texture images can be applied to the correction of original DCE-MRI without lesion distortion. The performance of the proposed registration method is evaluated on 18 lesions with obvious displacement by comparing the values of mutual information (MI), correlation coefficient (CC) and root mean square (RMS) before and after the registration. The experimental results show that the registration of original DCE-MRI based on the deformation field calculated by texture images is without lesion distortion. Compared with utilizing the original DCE-MRI and structure images, the classification result using corrected structure images based on TIC and feature selection is more accurate.