LUNG-GPT: Lung sound analysis with LLM-Based model

Lung diseases, such as Chronic Obstructive Pulmonary Disease (COPD), are becoming increasingly prevalent due to environmental pollution and unhealthy lifestyle choices, posing significant challenges for real-time monitoring and diagnosis. This study introduces Lung-GPT, a Generative Pre-Trained Transformer (GPT)-based model designed to analyze lung sounds collected from digital stethoscopes. Using data from the ICBHI challenge dataset and real-time measurements from the WEARME device, the model provides a comprehensive approach to disease classification and symptom detection. Key contributions of this research include the development of a Large Language Model (LLM)-based classifier capable of distinguishing between COPD, healthy, and other lung conditions, as well as segmentation models designed to detect and quantify wheezes and crackles in lung sound data. These models are further integrated with the WEARME device, enabling real-time detection and analysis of respiratory anomalies. The Lung-GPT model demonstrated robust performance, achieving a classification accuracy of up to 92% for COPD and effective segmentation outcomes. This work underscores the potential of transformer-based architectures in advancing lung sound analysis and monitoring solutions, with future improvements aimed at incorporating additional lung sound features for enhanced diagnostic precision.