- Tuberculosis is considered one of the deadliest dis- eases. It is caused by a Mycobacterium Tuberculosis. The life of the person becomes very severe if it is not treated at an early stage. The introduction of new technological methods, software and hardware devices, is encouraging researchers to come out with more powerful computer-aided methods for diagnosing TB at an early stage. However, the use of deep learning methods requires an ample amount of datasets. Because of the non- availability of enough volume of the dataset, in this work, we are using a generative adversarial neural network which has a special feature of synthesizing the images such that one cannot differentiate whether the image is original or fake. This study uses chest x-ray images of the infected persons, performing data pre- processing including resizing and converting them to grayscale and the input images are labeled. The input data consists of 4200 total images both healthy and unhealthy images. This work uses a deep learning-based technique, GAN. It consists of two sub models, the generator model and the discriminator model. The results were plotted using standard library functions. Classification accuracy of 96% was obtained, the precision of 96%, recall 79%, and 98% f1 score. This is observed to be the outstanding performance obtained compared to the similar kind of works in the literature.

Generative Adversarial Network, Tuberculosis, Convolution Neural Network, Mycobacterium, CT-Scan