HealthGuard: AI-powered Early Detection of Tuberculosis using Machine Learning

Early and accurate diagnosis of respiratory diseases remains a critical challenge, particularly in resource-constrained settings where access to advanced clinical testing is limited. This paper presents an AI-based diagnostic system that leverages cough audio signals for the automated detection of respiratory conditions such as tuberculosis (TB) and chronic obstructive pulmonary disease (COPD). The proposed system employs a deep learning pipeline in which raw cough recordings are pre-processed and converted into Mel-spectrogram representations, enabling effective extraction of time–frequency features. A convolutional neural network (CNN) architecture, enhanced with transfer learning using ResNet-18, is utilized to learn discriminative patterns from the audio features. Model performance is evaluated using standard metrics including accuracy, sensitivity, specificity, and area under the ROC curve (AUC). Experimental results demonstrate that the system achieves reliable classification performance, indicating its potential as a non-invasive, low-cost, and scalable screening tool. The proposed approach highlights the feasibility of using audio-based AI diagnostics to assist early disease detection and support clinical decision-making.