Novel deep learning-based optimization framework for the classification of respiratory diseases using lung sound...

Area of Science:

• Medical Informatics
• Artificial Intelligence
• Biomedical Signal Processing

Background:

• Traditional respiratory disease diagnosis relies on imaging, which can be costly and invasive.
• Lung sound analysis presents a non-invasive, cost-effective alternative for respiratory disease detection.
• Automated classification of respiratory conditions using lung sounds requires robust feature extraction and classification models.

Purpose of the Study:

• To propose and evaluate a deep learning-based optimization framework for automated respiratory disease classification using lung sound analysis.
• To enhance the accuracy and reliability of respiratory disease classification through advanced signal processing and machine learning techniques.
• To leverage the ICBHI 2017 Respiratory Sound Database for developing and validating the proposed classification model.

Main Methods:

• Audio signals were preprocessed using band-pass filtering and a Denoising Autoencoder (DAE) for feature extraction.
• An Enhanced Bidirectional Long Short-Term Memory (EBiLSTM) network with residual connections and regularization was used for classification.
• Hyperparameters were optimized using the Average and Subtraction-Based Optimizer (ASBO) to maximize classification accuracy.

Main Results:

• The proposed EBiLSTM-ASBO model demonstrated statistically significant improvements over baseline methods.
• Overall accuracy improved by up to 18.51%, with consistent gains in precision, recall, and Matthews Correlation Coefficient (MCC).
• Statistical hypothesis testing confirmed the robustness of the performance enhancements (p < 0.05).

Conclusions:

• The developed deep learning framework effectively captures disease-specific acoustic patterns in lung sounds.
• The proposed model offers a reliable approach for multi-class respiratory disease classification.
• This non-invasive method shows promise for improving the diagnosis of respiratory conditions.