A Research Of Hepatic Fibrosis Diagnosis Based Upon B-Scan Liver Images' Texture Analysis

With the improvement of living conditions, chronic liver diseases are now increasing. While hepatic fibrosis is the essential process of most chronic liver diseases converting to cirrhosis, the diagnosis using B-scan ultrasound is mainly dependent on the experiences of physicians. The objective of this thesis is to provide a computer-aided method for the diagnosis of hepatic fibrosis by applying texture analysis theory into B-scan liver images' analysis.By systematically comparing and analyzing recent years' methods of texture analysis for B-scan liver images, and understanding liver's pathological changes and texture features, this thesis proposed a framework to the image recognition of hepatic fibrosis' B-scan images. And it followed two steps:First step, texture features extraction: The texture features to differentiate normal and fibrosis liver images by choosing three methods of feature extraction, including Grey Level Co-occurrence Matrix Method, Power Spectrum Method, and Fractal Model Method, are based upon spatial domain, frequency domain, and algorithm model, respectively. Making use of their statistical difference between normal and fibrosis group, we finally set 6 feature descriptors: ASM, CON, HOM, ENE, VAR and PEL.Second step, texture auto-recognition: we used 40 samples (20 sampleseach) to distinguish these images according to a block of 64*64 pixels, inputting their 6 feature descriptors to a Back-Propagation Artificial Neural Network. The classification rate for fibrosis is 88%, which offers a satisfying performance for the computer-aided diagnostic system in future.