Article’s

Tuberculosis (TB) is still a major infectious disease which we are fighting at global scale and we require for it prompt and accurate diagnosis to breaks its transmission and to improve patient results. Presently we use traditional screening methods like sputum microscopy and manual interpretation of Chest X-rays (CXRs) which are very labor intensive, time consuming and also, we see great variation between readers which in turn affects the result especially in resource poor settings. In this work we present a full scale deep learning based system for what we have put forward is an automated TB detection solution also we aim to open up the AI which is at present a “black box”. We go in to the public chest x ray sets for preprocessing and then we use Deep Convolutional Neural Networks (CNNs) for very strong feature extraction and classification To solve the issue of standard AI models which are black box in nature we introduce Explainable AI (XAI) which we use Grad-CAM to visualized the infected areas. Also to close the gap between what we technically diagnose and what the patient understands we have put in a LLaMA-3 driven conversational AI which issues out medical reports in a human friendly language and also answers questions. We evaluate performance of the model to prove out its reliability. This study reports our work which is to present a transparent, intelligent and interactive tool for TB diagnosis via the use of this platform which in turn we hope will improve efficiency of clinical workflows