The Early Detection Of Liver Fibrosis And Classification Based On Ultrasonic Tissue Characterization

The quantitative analysis and even the classification research of the hepatic fibrosis has an important clinical significance for its early detection and prevention in cirrhosis. Based on the difference of microstructure of liver tissue is the target of early detection. This paper uses liver tissue in mice as experimental subjects and uses original ultrasound RF(Radio Frequency) time-series signal as data source. Based on ultrasound tissue characterization and correlation pattern recognition technology build five liver fibrosis recognition model and expect to provide a noninvasive clinical medicine effective means of early liver fibrosis monitoring and motion detection. The main research work of this thesis can be summarized as follows:1)By means of Hilbert transform for the RF signal display B-type figure and extract five texture features as well as six gray-level histogram characteristics from it. Based on the RF time series that extract 6 frequency-domain features, 5 time-domain features, 20 spectrum texture features and 25 time-domain texture features, analyze the distribution of each characteristic parameter during 5 stages, and provide an effective means of quantitative analysis for clinical ultrasonic analysis of the hepatic fibrosis.2)All the features are mixed together, and PCA and LDA are used for feature selection respectively, which reduce the model complexity by dimension reduction. Four kinds of normalization algorithms are used to normalize for features, which improve the robustness of the model. The recognition results are analyzed and compared by SVM and random forest. Experimental results show that when all features are mixed and the LDA feature selection algorithm is used and normalization algorithm is not used, random forest classification results is the best and accurate classification is up to 100%.3) A spectrum of texture can be combined with multi-field characteristics on liver fibrosis and effectively the five categories. Wherein the combination of textures and spectral image texture classification accuracy than conventional image texture based on the accuracy improved by more than 3%.4)Based on the MFC platform in VS 2010, making use of OpenCV, MySQL, openmp, Teechart and other relevant skills and knowledge to develop an application system which can realize the early detection and 5-level classification recognition of the hepatic fibrosis.