Clssification of nonalcoholic fatty liver ultrasonography by significant image features

The ultrasonography is widely used to diagnose nonalcoholic fatty liver disease (NAFLD). However, the diagnostic reports were affected by operating bias of interobserver and intraobserver. The main aim is to build a feasible classified model for NAFLD with ultrasonography images. The significant features of image are inner-quartile range (IQR), standard deviation (STD), and hepatorenal index (HI) of specific region of interest (ROI). A logistic regression classifier was a feasible classified model with predictors IQR, STD and HI. The accuracy, sensitivity, specificity, the area under receiver operating characteristic (ROC) curve (AUC), positive predictive value (PPV), negative predictive value (NPV), and kappa value were 0.88, 0.92, 0.93, 0.79, 0.92, 0.79, and 0.71 respectively. The presented approach provides high agreement classifier with physician diagnostic judgment.