Computer-aided Quantification Staging Of Liver Fibrosis Based On B-mode Ultrasound Images

Liver fibrosis is one of diseases which are seriously harmful to human health.Currently, B-mode ultrasound is a common method for examination of liver fibrosis.However, B-ultrasound could only diagnosis liver fibrosis qualitatively, but notquantitatively. Quantitative liver fibrosis staging system could depend on computer toanalysis the texture of B-mode images, extracting the information which is notobserved by human eyes. It can effectively assist doctors to diagnose liver fibrosis andimprove the efficiency and accuracy of clinical diagnosis.Liver fibrosis computer-aided diagnosis based on B-mode ultrasound imagesmainly consists of four processing steps: region of interest selection, featureextraction, feature selection, classifier training and prediction. Firstly, an interactiveapproach is proposed to select region of interest. Secondly, accordingto liver fibrosis in the characterization of B-mode ultrasound images, four methods oftexture feature extraction have been selected to distinguish the severity ofliver fibrosis, including general statistics, gray level co-occurrence matrix, spotoperator, gaussian markov random, time-frequency domain transform. Thirdly,correlation-based feature selection is used to filter the extracted texture features inorder to obtain effectively feature vector. Finally, the feature vector is input tosupport vector machine for training and recognition.36patients with chronic hepatitis C were investigated using B-mode ultrasoundand liver biopsy in Beijing Friendship Hospital. We selected96region of interest from49B-mode ultrasound images to carry out experiments. The result showed that thebest classification accuracy of no liver fibrosis, mild fibrosis and severe liver fibrosiswere96.9%,78.1%and84.4%, respectively. The experiments revealed that,quantitative liver fibrosis staging system could certainly provide clinical diagnosticinformation, assisting doctors to diagnose.