Deep Learning Based Tuberculosis Classification from Chest X-ray Images

Tuberculosis (TB) is a serious health issue that kills a lot of people these days. TB is completely curable if it is discovered early enough. One way of diagnosing tuberculosis (TB) early on is to undergo a chest X-ray (CXR), which can show if a person has active TB. Chest x-ray images (CXR) can be analyzed to diagnose the deadly TB disease more accurately owing to deep learning (DL), a significant area of artificial intelligence. To automate tuberculosis (TB) identification, this study used Convolutional Neural Networks (CNNs), a form of deep learning models designed specifically for image processing. This architecture needs a lot of datasets to learn features, which makes the learning process take longer. Transfer Learning (TL), which uses a large number of pre-trained models to perform disease classification, is suggested as a solution to this issue. In this work, four different CNN architecturesEfficientNetB0, AlexNet, GoogleNet, and Xception-are trained and evaluated. The results demonstrated that GoogleNet could diagnose tuberculosis with an astounding 97% accuracy, even after extended training periods. By comparison, AlexNet trained faster due to its simpler architecture, but having a little lower accuracy of 96%.