Teleconsultation-Enabled Pulmonary Assessment Using a Multi-Task Deep Learning Framework

Abstract

Accurate assessment of pulmonary function is essential in healthcare for the timely diagnosis and management of respiratory conditions. Conventional studies have predominantly analyzed lung sounds and diseases as separate tasks. This study introduces a customized multi-task deep learning framework capable of performing simultaneous classification of lung sounds and lung diseases. The ICBHI 2017 Respiratory Sound Dataset was utilized for model development, wherein Mel Frequency Cepstral Coefficients (MFCCs) were extracted from lung sound recordings and concatenated with corresponding labels for training, validation, and testing. The proposed model achieved weighted average precision, recall, and F1-scores of 88%, 89%, and 86% for lung disease classification, and 61%, 65%, and 58% for lung sound classification, respectively. The overall classification accuracies were 89% for lung diseases and 65% for lung sounds. The trained model is integrated within a cloud-based health informatics framework that enables bidirectional data exchange via a dedicated mobile application. Patients can upload demographic and respiratory sound data to the cloud, where automated classification is performed, and results are made accessible to healthcare professionals. Clinicians can subsequently provide therapeutic recommendations through the same platform. The proposed system demonstrates the potential for real-time deployment in clinical and telehealth environments, promoting accessible and intelligent respiratory health monitoring.