AI-Based Chest Disease Detection System

Authors: Prachi D. Vighne, Trupti A. Dalimbe, Dnyaneshwari M. Khedekar, Sudarshan J. Sikchi

Abstract: Chest diseases such as pneumonia, tuberculosis, and COVID-19 continue to be major global health concerns, leading to significant morbidity and mortality. Accurate and early diagnosis is essential for effective treatment; however, traditional methods rely heavily on expert radiologists to interpret chest X-ray images, which can be time-consuming and subject to variability. In many regions, the shortage of skilled professionals further limits timely diagnosis, highlighting the need for automated and reliable solutions. This work presents an Adaptive Explainable AI framework for chest disease detection that combines deep learning, interpretability, and confidence estimation. A Convolutional Neural Network (CNN) is utilized to automatically learn relevant features from chest X-ray images and classify them into multiple categories, including COVID-19, pneumonia, tuberculosis, and normal cases. To address the lack of transparency in deep learning models, explainable AI techniques such as Grad-CAM and saliency maps are incorporated. These methods provide visual insights by highlighting regions in the X-ray image that contribute most to the model’s prediction. In addition, the system estimates prediction confidence using probability-based measures, allowing users to assess the reliability of the output. A severity estimation module is also included, which analyzes the extent of affected regions in the image to categorize the condition into levels such as mild, moderate, or severe. This adds practical value for decision-making and prioritization. The proposed system is implemented as a web-based application, enabling users to upload chest X-ray images and receive real-time predictions along with visual explanations and severity assessment. Experimental observations indicate that the model achieves satisfactory performance while improving interpretability and user trust. Overall, the framework provides a balanced approach between accuracy, transparency, and usability, making it suitable as a supportive tool in medical diagnosis.