The Pilot Study Of Computer-Aided Diagnosis Of Hepatic Fibrosis

Objective By using the software of computer-aided diagnosis to analyze the hepatic texture on the images of conventional computer tomography, the correlation of texture features to chronic hepatopathy and pathologic grades is to be found for the purpose of the diagnosis of hepatic fibrosis in its early stage.Materials and methods 146 patients who underwent percutaneous liver biopsy and intraoperative biopsy,who did not undergo pathological examination but have no hepatitis history or normal and who have typically clinical and radiological changes were performed with the multi-detector 64 row CT scanner, including 36 patients who were normal,39 patients who were mild fibrosis(S1+S2)?35 patients who were sever fibrosis(S3+S4)?36 patients who were liver cirrhosis. Texture features were calculated by the computer algorithm form the selected ROI on the liver parenchymal. With the supporting of Matlab software platform, the texture features from the Liver Cirrhosis 4.8 software were inputted to a support vector machine (S VM) algorithm so as to build a liver fibrosis model. The structure of SVM is composed of a seven input layer, which are contrast angular second moment?entropy?correlation% inverse difference moment?mean gray value and standard deviation. While four output layers are normal(SO)?mild fibrosis(S1+S2)?sever fibrosis(S3+S4)?liver cirrhosis. The leave-one-out method is used to evaluate the accuracy of each model.Results Firstly, we build a model to differentiate normal liver from liver cirrhosis for choosing the best size thickness, and structure of datasets. Two of the best datasets are the original data with texture features calculated from no filter?ROI 20×20 (pixels), that are evaluated as the highest accuracy by leave-one-out method. The accuracy rate is 85.41%?with the sensitivity 86.67% and the specificity 84.38%. The others are using the feature texture of different kind of thickness of layer, such as 0.625mm?2.5mm?3.75mm, to evaluate normal and liver cirrhosis. Their accuracy are 89.07%?87.15%? 89.71%. Secondly, liver fibrosis is evaluated by using the best dataset above. In two classifiers, the model built by the best dataset selected from 3.75mm CT images has the highest accuracy in normal and liver cirrhosis. The accuracy rating is 86.06 with the sensitivity 87.77% and the specificity 89.06%. This dataset is also evaluated with the highest accuracy in mild and sever fibrosis, both accuracy rate are 73.53%. In four grad classification, the model did not work well. The highest accuracy rate was from the model built by the best dataset in 2.5mm CT images with accuracy rate 50.00%.Conclusions Texture features analyzed by Computer algorithm on hepatic CT images are used in building model. SVM method is able to build liver fibrosis and cirrhosis model via studying limit samples. The result demonstrates that it is correlated the degree of fibrosis accurately. Texture analysis using the computer algorithm on CT image is found to have a potential usefulness for the diagnosis of hepatic fibrosis.