Computer-Aided Diagnosis for Pneumoconiosis Based on Texture Analysis on Digital Chest Radiographs

Abstract

In this paper, texture analysis was used to discriminate digital chest radiographs of pneumoconiosis patients from normal ones. First, lung fields in each chest radiograph were segmented by using the morphological reconstruction and Otsu-thresholding. Second, several texture features based on the histogram and co-occurrence matrix of grey levels were extracted. Finally, a neural network based classifier was trained with features extracted from 66 chest images to distinguish pneumoconiosis patients from normal cases. Another 29 images were used to assess the diagnosis performance of the classifier, giving an overall accuracy of 79.3%.