Abstract: A new computer aided diagnosis (CAD) method for hepatocirrhosis is proposed, based on the recognition of characteristics of liver texture and shape. In the method, the liver region is segmented from the abdominal MRI-based image through image segmentation technique, and the characteristics of liver texture and shape are extracted from liver region. Then, the degree of liver fibrosis is calculated according to its texture and the improved BP neural network by conjugate gradient algorithm, and the passivation of left liver leaf is calculated according to its shape characteristics. Finally, the hepatocirrhosis is diagnosed taking the degrees of liver fibrosis and passivation as criteria. The method has been applied to the practical abdominal MRI, and showed higher accuracy in the diagnosis of hepatocirrhosis than single neural network.