Acoustic-Based Deep Learning Architectures for Lung Disease Diagnosis: A Comprehensive Overview

Area of Science:

• Respiratory Medicine
• Artificial Intelligence
• Biomedical Signal Processing

Background:

• Lung auscultation is a vital diagnostic tool for respiratory health, gaining attention post-pandemic.
• Computer-based respiratory sound analysis, particularly using AI, offers advanced methods for detecting lung abnormalities.
• Existing reviews lack specific focus on deep learning architectures for lung sound analysis.

Purpose of the Study:

• To provide a comprehensive review of deep learning-based architectures for lung sound analysis.
• To consolidate information on trends, datasets, features, and methods in this specialized field.
• To identify gaps and suggest future research directions in AI-driven respiratory diagnostics.

Main Methods:

• Systematic literature search across major scientific databases (PubMed, IEEE, Springer, etc.).
• Extraction and assessment of over 160 publications on deep learning and lung sound analysis.
• Analysis of common features, datasets, classification methods, and signal processing techniques.

Main Results:

• Identified key trends in lung sound pathology and classification.
• Summarized common features and datasets used in deep learning models for respiratory sounds.
• Highlighted various signal processing techniques and classification methods employed.

Conclusions:

• Deep learning offers significant potential for advancing computer-based respiratory sound analysis.
• Further research is needed to refine models, improve data standardization, and explore novel architectures.
• This review provides a foundation for future development in AI-assisted lung sound diagnostics.