Quantitative analysis and identification of liver B-scan ultrasonic image based on BP neural network

This paper aims to provide a method of quantitative analysis and intelligent identification for diagnosing fatty liver by B-scan ultrasonic image. By analyzing textural features of liver B-scan ultrasonic image, some features including angular second moment, entropy and inverse differential moment of gray level co-occurrence matrix are extracted from B-scan ultrasonic liver images, and then feature vectors are formed, which are input BP neural network for training classification and identification. BP neural network classifier can identify normal liver and fatty liver with the accuracy rate 85.71% and 88.89% respectively, at the same time, can identify slight, moderate and serious fatty liver with the accuracy rate 71.42%, 71.42% and 85.71% respectively. The method proposed here for quantitative analysis and diagnosing fatty liver can provide reference for doctor clinical diagnosis, and it will greatly improve the accuracy and efficiency of the fatty liver diagnosis combining clinical experience.