Computer Aided Detection for Pneumoconiosis Based on Co-Occurrence Matrices Analysis

Abstract

This paper presents a texture analysis method on digital chest radiograph to distinguish pneumoconiosis chest from normal chest. First, two lung fields are segmented from a digital chest X-ray image by the active shape model (ASM) method and regions of interest (ROIs) are selected in inter-rib areas along the outer and middle zones of the lung fields. Second, the chest image is preprocessed by multi-scale difference filter bank to enhance some detailed features of pneumoconiosis. Then the co-occurrence matrices features are extracted from each ROI, including energy, entropy, local homogeneity, correlation and inertia. A support vector machine (SVM) classifier is utilized here to extract the discriminatory information through leave-one-out cross validation. The analysis result based on the database with ground truth shows that normal regions could be differentiated from abnormal regions distinctively. The prediction classification performance on the manual ROIs database has sensitivity 95.6%, specificity 94.2%, and the overall accuracy 95.15%.