Computer Aided Diagnosis of Fatty Liver Ultrasonic Images Based on Support Vector Machine
Guokuan Li,Yu Luo,Wei Deng,Xiangyang Xu,Aihua Liu,Enmin Song
DOI: https://doi.org/10.1109/iembs.2008.4650279
2008-01-01
Abstract:B-scan ultrasound is the primary means for the diagnosis of fatty liver. However, due to use of various ultrasound equipments, poor quality of ultrasonic images and physical differences of patients, fatty liver diagnosis is mainly qualitative, and often depends on the subjective judgment of technicians and doctors. Therefore, computer-aided feature extraction and quantitative analysis of liver B-scan ultrasonic images will help to improve clinical diagnostic accuracy, repeatability and efficiency, and could provide a measure for severity of hepatic steatosis. This paper proposed a novel method of fatty liver diagnosis based on liver B-mode ultrasonic images using support vector machine (SVM). Fatty liver diagnosis was transformed into a pattern recognition problem of liver ultrasound image features. According to the different characteristics of fatty liver and healthy liver, important image features were extracted and selected to distinguish between the two categories. These features could be represented by near-field light-spot density, near-far-field grayscale ratio, grayscale co-occurrence matrix, and neighborhood gray-tone difference matrix (NGTDM). A SVM classifier was modeled and trained using the clinical ultrasound images of both fatty liver and normal liver. It was then exploited to classify normal and fatty livers, achieving a high recognition rate. The diagnostic results are satisfactorily consistent with those made by doctors. This method could be used for computer-aided diagnosis of fatty liver, and help doctors identify the fatty liver ultrasonic images rapidly, objectively and accurately.