Ultrasound imaging is an important tool for early detection and regular check-ups of liver cirrhosis. The diagnosis can be performed by analysis of echo textures of the liver and of the accompanying spleen. The simultaneous comparison of liver and spleen images for the same person at the same system setup can be used to reduce subject, machine, and system variations. This study aims to investigate the computer-aided diagnosis of features derived from the ultrasound images of livers and the accompanying spleens. We will incorporate the techniques of an early vision model, dimension reduction, fractal dimension, nonparametric discriminant rules by kernel density estimation and classification trees to improve the statistical analysis methods. These methods are tested by the clinical images collected at National Taiwan University Hospital with 64 normal livers and 30 cirrhosis ones. The smallest overall bootstrap prediction error is found to be 5.29% by these new methods.