Computer-Aided Diagnosis Of Hepatic Fibrosis On Image Shape Feature

Hepatic fibrosis is a result of proliferation and deposition in numerous liver tissues. The structure of hepatic lobule will be destroyed in consequence of the fibrosis. Hepatic fibrosis is an essential process of liver cirrhosis and research found that the process is reversible. It is significant to be detected earlier and treated timely to prevent liver cirrhosis.In modern medical research, X-CT technology, Magnetic Resonance Image (MRI) and Ultrasonic Imaging are employed to obtain clear and intuitive medical images with the help of digital medical image processing technology. The accuracy for the diagnosis regarding to internal pathological changes and their reasons have been improved gradually. However, diagnosis of hepatic fibrosis on medical images relies mainly on doctor’s perusal so these medical images are underutilized. In order to improve the utilization of medical images and the accuracy of diagnosis, the primary coverage of the paper is arranged:(1) Conversion of DICOM to BMP image. Sobel, a traditional edge detection operator, is applied to detect the contour feature of liver in BMP image. The contour is also extracted manually in the DICOM image. And compare about their differences.(2) Dispose of CT image contour in 2D coordinates. According to the waviness characteristic parameters of micro mechanical surface,10 surface shape features are extracted in CT liver contour. Then conduct diagnosis experiments on classifying the degree level of hepatic fibrosis.(3) Classification and diagnosis experiment for quantitative analysis with Support Vector Machine (SVM). Cyclic traversal method and leave-one-out method are applied to exhaust combination of all features which will be check in fibrosis degree level. Control variable method is employed to analyze the results. It is shown that the optimal number of features is confirmed from 2 to 6 considering the results of experiment. Ranking the weight of every shape feature in the experiment and some data are obtained, of which Rq, Rmax, Rp, Rm and D weight heavier. Applying each shape feature to other classification experiments and analyzing the gradient change for the accuracy of these experiments.(4) Classification performance between shape feature and texture feature.10 shape features and 15 texture features are combined together to apply to classification and diagnosis experiment. Analyze effective new features and value of weight with the old combined features.