Abstract
The most common lung diseases, such as pneumonia and tuberculosis, remain some of the biggest global health diagnosis challenges, particularly in areas with extremely poor access to expert radiologists. Chest X- rays are affordable and accessible; however, their interpretation requires a great deal of expertise, which might not be consistently available across different clinicians. Recent advances in artificial intelligence, particularly deep learning, present promising solutions by automating and improving the interpretation of radiographic images. This study presents a robust system contributing to improved diagnostic performance by processing chest radiographs using state-of-the-art deep learning techniques. Various models were trained and evaluated for the detection of tuberculosis and pneumonia. An individual best performing model could achieve an accuracy of 98.73% while the result after an ensemble of diverse deep models could achieve a test accuracy of 98.05%. That proves that diverse deep learning models can substantially improve medical image analysis, enabling the development of more reliable diagnostic tools and offering accessibility across high- resource and low-resource healthcare settings. Code and all models have been made publicly available to foster transparency and subsequent research in AI-driven medical diagnostics.