SD Rats' Fatty Liver Tissue Classification Based on Ultrasound Radiofrequency Signal

摘要

Early diagnostic of mild fatty liver and prediction for the level of fatty liver have important value in clinic. However, it's difficult to diagnose mild fatty liver and grade fatty liver correctly using ultrasonic images. The method based on extracting features from ultrasound radiofrequency (RF) signals was proposed in this paper. Five features were selected: msr(mean/sd), skewness and kurtosis of RF signal envelope, the approximate entropy of signal and the energy ratio of two regional signals. Then back-propagation(BP) network combined with "leave-one-out" cross-validation was employed to classify normal livers and variable degree of fatty livers. The results showed that, the features could distinguish normal livers from pathological livers with accuracy rates of 98%. The accuracy rates of classification were 98.08%, 82.05%, 78.38%, 87.5% for normal liver, mild fatty liver, moderate group and severe group, separately.