3457: Sonographic texture analysis for the computer-aided diagnosis of the liver cirrhosis

Abstract

Objectives: Various statistical classifiers including the textural parameters are tried for the computer-aided diagnosis of the liver cirrhosis using the ultrasound (US) images. Methods: The gray scale B-mode US images of 93 patients with/without liver cirrhosis were gathered and diagnosed by two experienced radiologists. The textural characteristics of the liver and spleen parenchyma is computed and compared with the clinical diagnosis. The representative coarseness levels (RCLs) were determined from the various statistical parameters linearly combined with the coefficients by the Levenberg-Marquardt minimization in order to improve the correlation with the clinical results. Results: The classification among clinical diagnostic results of the liver cirrhosis is performed using the RCLs representing the US hepatolienal textural contrast. The obtained RCLs are correlated fairly well with the clinical diagnostic results, and the false-negative rate is obtained to be about 7%. Receiver operating characteristics analysis reveals Az = 0.96 showing their effectiveness for the diagnosis. Conclusions: The textural characteristics of US abdominal images have been investigated for the computer-assisted diagnosis of the liver cirrhosis giving the promising results for the computer-assisted diagnosis of the liver cirrhosis.