Characteristic Information Analysis Of Fatty Liver And The Aided Diagnosis System Design Based On Ultrasonic RF Signal

Fatty liver is a very common hepatic disease. Currently, the regular inspection anddiagnosis of fatty liver basically rely on ultrasound technology in the clinical. There isobviously insufficient on the objectivity of ultrasonic diagnosis of fatty liver, especiallyin the quantitative diagnosis for fatty liver. So, improving the ability of ultrasonicquantitative diagnosis for fatty liver has great clinical value.In response to these circumstances, the thesis study the methods of improving theability of diagnosis in ultrasonic images and quantitative diagnosing fatty liver byanalyzing the characteristics information in the radio frequency(RF) signal of fatty liver,then design a computer-aided diagnosis system for diagnosis of fatty liver. The maincontent of the thesis are as follows:1. Analyzing the characteristic information in the RF signal of fatty liver.Researching the ultrasonic imaging effects with different methods of processing RFsignal by using digital signal processing techniques. Looking for the characteristicsignal between the normal liver and different severity fatty liver.2. Reconstructing the ultrasound images which have better diagnostic capabilitiesfor fatty liver. By analyzing the characteristics of fatty liver in B-mode images, amethod of processing image with piecewise linear transformation in grayscale isproposed. The method improves the ability of image for diagnosing fatty liver.3. Researching the method for quantitative diagnosis of fatty liver. By analyzingthe screening fatty liver characteristic parameters, we determine the best feature vector,including the mean center frequency ratio, the mean amplitude ratio, the ratio of meanand standard deviation, kurtosis and skewness. Then employing the BP neural networkto classify the normal liver and fatty liver in order to achieve the quantitative diagnosisof fatty liver.4. Designing the computer-aided diagnosis system in the VC6.0developmentplatform. Designing the ultrasounic signal processing modules based on the RF signal toreconstruct and display the B-mode image of liver. Then achieving the classification andrecognition of fatty liver by building BP neural network. In this study, we do experiment with hepatic ultrasonic RF signal acquiring fromWest China Hospital. There are86groups hepatic RF data, including28samples withnormal liver,22samples with mild fatty liver,20samples with moderate fatty liver,16samples with severe fatty liver. The accuracy rates using the computer-aided diagnosissystem to classify the normal liver and fatty liver were96.4%and98.3%. And theaccuracy rates of classification were86.4%,90%,93.8%for mild fatty liver, moderatefatty liver and severe fatty liver, separately. The results show that the diagnosis systemdesigned in the thesis has a certain clinical value in diagnosing fatty liver.