Texture-Based Statistical Detection and Discrimination of Some Respiratory Diseases Using Chest Radiograph

Abstract

This chapter proposes a novel texture-based statistical procedure to detect and discriminate lobar pneumonia, pulmonary tuberculosis (PTB), and lung cancer simultaneously using digitized chest radiographs. A modified principal component method applied to wavelet texture measures yielded feature vectors for the statistical discrimination procedure. The procedure initially discriminated between a particular disease and the normals. The maximum column sum energy texture measure yielded 98 % correct classification rates for all three diseases. The diseases were then compared pair-wise, and the combination of mean of energy and maximum value texture measures gave correct classification rates of 70, 97, and 79 % for pneumonia, PTB, and lung cancer, respectively.

Computer-aided Tuberculosis Diagnosis with Attribute Reasoning Assistance

“…Accurately recognizing and localizing diseases in chest X-Ray images is very challenging because of low textural contrast, large anatomic variation across patients, and organ overlapping. Previous works in this field mainly focus on disease classification [17,24,1,27]. Recently, the authors in [4] propose to transfer deep models pretrained on the ImageNet dataset [5] for recognizing pneumonia in chest X-ray images.…”