Hierarchical Deep Learning for Multi-Stage Lung Disease Classification in Chest X-rays

Lung diseases continue to be a serious public health concern that require prompt and precise diagnosis in order to facilitate efficient treatment and lower mortality. An AI-powered diagnostic tool designed especially for lung disease detection in the Indian healthcare system is presented in this research project. We offer a reliable and scalable method that uses cutting-edge deep learning techniques to automate the diagnosis of four main lung diseases: pneumonia (Bacterial and Viral), COVID-19, tuberculosis, and normal. We combine the best features of DenseNet for the model architecture. DenseNet's densely connected layers improve the model's capacity to identify minute anomalies by facilitating rich feature extraction and effective gradient flow. All four classes classification performance is enhanced by the Densnet adaptation for accurate image-to-class mapping. For transparency and clinical trust, techniques like Grad-CAM are used to highlight the regions that have the greatest influence on the AI's decisions. The model exhibits good accuracy, sensitivity, and specificity in differentiating between healthy and diseased lungs. By providing quick and accurate initial evaluations, this AI tool has great potential to assist radiologists, especially in rural or resource-constrained areas. It seeks to aid in early identification, lessen diagnostic workload, and enhance patient outcomes in various medical settings.