Research Of Liver Steatosis Degree Recognition Based On Ultrasonic Medical Image

With the improvement of the living conditions, people pay much more attention to their health. Although fatty liver is a benign lesion in pathology, it's also a very important index of a body's "subhealth". If it is left untreated, fatty liver may be associated with or may lead to inflammation, fibrosis, cirrhosis and possibly cancer. The B-mode ultrasonography is widely used in diagnosing fatty liver. But it mostly relies on the doctor's observations and experiences. As a result, it is subjective and may be unreliable. With the development of texture analysis technology, it is possible to diagnose fatty liver of B-mode ultrasonic image by computer-aided diagnosis. There are some researches about the recognition of the normal liver and the fatty liver. But researches about the liver steatosis degree recognition based on ultrasonic medical image are not available yet. Therefore, the aim of this paper is to provide a computer-aided method for the diagnosis of liver steatosis degree by B-mode ultrasonic imaging.This study focuses on the recognition techniques of the fatty liver ultrasonic image based on texture analysis. Firstly, B-mode ultrasonic images are collected from Huaxi Hospital of Sichuan University. Secondly, the regions of interest are localized. During the feature extraction procedure, the mean intensity ratio of the near field and the far field (MIR), the angular second moment based on wavelet transform(ASM), the singularity strength width(SSW), and the multi-spectrum area(MSA) are extracted. Then, C-means clustering algorithm and back-propagation (BP) artificial neural network are employed respectively to classify these feature vectors.There are totally 224 patients studied in the experiment. Their liver images are classified into four groups: normal, mild, moderate, and severe fatty liver. Each of these groups has 56 patients. By different combinations of MIR, ASM, SSW, and MSA, the images can be classified into four classes. Firstly, By using C-means clustering algorithm, we get the best feature vector including MIR, SSW, MSA, which has the best classification result. Secondly, the best feature vector is computed by using the BP neural network. The accuracy rates with BP neural network are 96% for normal liver, 80% for mild fatty liver, 88% for moderate fatty liver, and 92% for severe fatty liver.The result shows that the MIR, the MSA and the SSW can describe the features of B-scan ultrasonic liver images successfully. The supervised approach by BP neural network has good performance in identifying the images of liver with different degree of steatosis. All of the above features are effective in diagnosing liver fatty degree by B-mode ultrasonography.