Chest X-ray Image Classification Using Convolutional Neural Network to Identify Tuberculosis

Abstract

Tuberculosis (TB) is a severe bacterial infection that can be spread by inhaling small droplets from an infected person’s cough and sneeze. TB claimed the lives of 1.5 million individuals in 2020, including 214,000 HIV-positive people. TB is the world’s second most widespread infectious lethal disease and the 13th leading cause of mortality. As a result, predicting whether someone has tuberculosis or not is critical. We experimented with chest X-ray images of healthy and tuberculosis patients. For our studies, we applied the CNN models VGG16, VGG19, Xception, ResNet50, InceptionResNetV2, DenseNet201, InceptionV3, and MobileNetV2. We also developed two models utilizing convolutional layers, max-pooling, and other techniques. In our study, VGG-16, Xception, and DenseNet201 provide any model’s highest training and validation accuracy. Densenet201 has the highest accuracy, with 99.7% in validation and 99.7% in training. One model we have developed has good training and validation accuracy, with 90.7% in training and 90% in validation.