Computer Aided Diagnosis Of Breast Tumor Based On Sequential Grayscale Ultrasound Images

Nowadays, sonography is one of the most frequently used methods for the early detection of breast tumor. A large amount of research shows that, the accuracy of detection using the sonography can be further improved by combining the computer aided diagnosis (CAD) technology. The breast tumor CAD system usually performs the diagnosis based on the supplementary information of morphology and grayscale texture characteristics. However, the performance of the CAD system utilizing the aforementioned two characteristics is still not so satisfactory to date. Clinical research has demonstrated that the tumor elasticity is also a very important indicator to judge whether the breast tumor is benign or not. This dissertation studied the performance of a breast tumor CAD system with the particular inclusion of the tissue elasticity characteristics, which were obtained by sequential grayscale ultrasound images collected during a tissue compression process. According to the conventional processes of a CAD system, the main research work and contribution of this dissertation can be summarized as follows:(1) Preprocessing the sequential ultrasound images. A filtering algorithm which utilized the correlation information in space, grayscale and time domains simultaneously was proposed for the video-format sequential ultrasound images. In order to improve the stability of the algorithm, the Gaussian weighting kernel was adapted in all the three domains to reduce the sensitivity of the filtering results to the selected threshold. Furthermore, an improved anisotropic diffusion speckle reduce algorithm was proposed to deal with the single images with severely high noise. Through mitigating the over-saturation problem of the diffusion coefficients in the original algorithm and the deficiency in choosing the speckle scale coefficients, the impact of human factor effect was reduced, and the stability of the new algorithm was enhanced.(2) Segmenting the ultrasound images of breast tumor to get the tumor boundary. This dissertation proposed an improved C-V model, which could avoid the step of re-initialization, thus the speed of segmentation being accelerated greatly. Furthermore, based on the grayscale distribution characteristics of the breast tumor ultrasound images and the hypothesis of a piecewise constant in the C-V model, a semiautomatic segmentation flow was presented, in which the rough contour was sketched first, and then a subimage was obtained to apply the refined segmentation algorithm. This flow improved not only the accuracy, but also the efficiency of the segmentation algorithm. For the sequential images, using the segmentation result of each frame as the initial boundary of the next frame, can reduce the impact of human factors and enhance the segmentation efficiency. In addition, through analyzing the deficiency of the existing evaluation methods for the medical image segmentation, a new evaluation method was proposed, and then the segmentation results of breast tumor ultrasound images were evaluated by using both the existing methods and our new method.(3) Extracting the characteristic parameters of the breast tumor from the segmentation results. Three classes of parameters, include morphology, grayscale and elasticity characteristics, were extracted in this study, in which 12 morphological characteristics and 3 grayscale characteristics were proposed by others. Considering that the freehand compression process was very difficult to keep a constant speed, an evaluation algorithm was proposed to quantify the compression depth between two consecutive images in the image sequence. On this basis, 10 elasticity parameters were proposed to describe the extent of deformation under per unit compression depth. Among the elasticity parameters, two characteristic parameters named total deformation and shrink-magnify ratio were calculated by the method of nonrigid registration deformation field in a novel way.(4) Using the characteristic parameters that were extracted in the previous step to aid the classification of breast tumor. This dissertation selected the best combination of parameters by a manual process. Some best combinations of the elasticity parameters were selected first as the basic combinations and then morphological and grayscale parameters were added in to test the overall performance. In the testing process, both the support vector classification and the support vector regression were employed to analyze the data, and three indicators named extreme distance, mean distance and inter-class distance were proposed to compare the performance of different characteristic parameter combinations.Based on the above four steps, 187 pathologically proven cases including 85 malignant tumors and 102 benign ones were tested in this study. The experiment result showed that the performance of the CAD system which used the morphology, grayscale and elasticity characteristics combinatively. was much better than the system which only used the morphology and grayscale characteristics. Therefore, it was concluded that the elasticity characteristics based on the sequential grayscale ultrasound images performed well in the classification of breast tumors, and it could be used as supplementary information for the diagnosis of breast tumors.