Enhanced TB Detection to Solve Class Imbalance and Explainability using GenAI and Deep Learning Technique

One of the leading causes of infectious disease mortality in the world is tuberculosis. Although early screening using chest X-ray images is important, it is often limited by the availability of radiologists and class imbalance in real-world datasets where the number of samples with tuberculosis is much lower compared to normal cases. With the purpose of addressing class imbalance and improving explainability, this paper proposes an enhanced framework for tuberculosis detection that couples a lightweight deep learning model with generative artificial intelligence. In order to balance the dataset, a StyleGAN2 model was utilized to generate synthetic chest X-ray images. The enriched data served to train a MobileNetV2 classifier from scratch. Explainability was achieved through Grad-CAM++ and Guided Backpropagation, allowing the clear visualization of model decision regions. The experimental results obtained demonstrate significant gains in minority-class recall, overall accuracy, and interpretability.