Tuberculosis (TB) classification in chest radiographs using deep convolutional neural networks

Tuberculosis (TB) is a deadly disease and a major health threat in most of the low and middle income countries in the world. Chest radiography is the most common method for TB screening. The success of this type of analysis depends on the experience and interpretation skill of the radiologists who examine the chest radiographs (chest X-Rays). The development of computer-aided diagnostic (CAD) systems has accelerated the early diagnosis of TB presentations. Within the field of medical imaging in recent years, Deep Convolutional Neural Networks (CNNs) have played a vital role in segmentation techniques and feature extraction for disease detection and classification of X-rays as normal or abnormal. This paper describes the development of a simple CNN architecture, with image augmentation and X-Ray image pre-processing, to classify chest radiographs into TB positive and TB negative classes. The performance of the system is measured on three publicly available datasets: the Shenzhen chest X-ray set, the Montgomery County chest X-ray (MC) set, and the India chest X-ray set. The proposed computer-aided diagnostic system for TB screening, achieves accuracy of 87.29%, which is comparable to the performance of human radiologists.