Hybrid CNN-LSTM with Generative AI for Classification of Respiratory Diseases Using Lung audio Sound

Abstract

Respiratory diseases such as asthma, chronic obstructive pulmonary disease (COPD), lung cancer, and tuberculosis pose significant global health challenges. Accurate and efficient classification of these conditions is vital for improving patient care and optimizing healthcare resources. This study presents a hybrid deep learning model that integrates Convolutional Neural Networks (CNNs) and Long Short-Term Memory (LSTM) networks to enhance lung sound analysis for diagnosing respiratory diseases. The proposed system follows a structured approach comprising three key stages: preprocessing, feature extraction, and classification. In the preprocessing stage, lung sound recordings undergo resampling, noise reduction, segmentation, and augmentation to improve data quality. Generative Adversarial Networks (GANs) are employed to address data scarcity by synthesizing realistic lung sound samples. Feature extraction is performed using log-scaled mel spectrograms, capturing both spectral and temporal information essential for identifying respiratory patterns. The classification model leverages CNNs for spatial feature learning and LSTMs for capturing sequential dependencies, resulting in a high classification accuracy of 99.6%, surpassing conventional CNN-based approaches. Additionally, the system incorporates explainability techniques, such as Gradient-weighted Class Activation Mapping (Grad-CAM), to highlight significant spectral features influencing predictions, enhancing transparency and aiding clinical validation. By automating respiratory disease detection, this approach enables rapid, cost-effective, and non-invasive screening, reducing the dependence on specialized medical expertise, particularly in resource-limited healthcare settings. The proposed method aligns with clinical standards, contributing to early diagnosis and improved disease management