Differentiation of cirrhosis from normal liver based on textural features via T1WI computer-aided diagnosis with a genetic algorithm

摘要

A computer-aided diagnosis (CAD) system for classification of liver cirrhosis from MRI is presented. The system consists of feature extraction and selection, classification, and classifier optimization modules. In general, biomedical imaging is based on textural features, visualized via grey level co-occurrence matrices. However, these features are so numerous that it is difficult to determine which are the most effective for classification. Then feature selection was facilitated by application of a box plot. In addition to ensure the stability of the back-propagation (BP) classifier and improve its performance, a genetic algorithm (GA) was incorporated. We demonstrated that the proposed CAD system is suitable for differentiation through analysis of 170 regions of interest in T1WIs of advanced cirrhosis and normal livers. The GA improved classification performance of the BP classifier, allowing fewer iterations, less time expense, and a high accuracy rate.