Abnormal Respiratory Sound Identification Using Audio-Spectrogram Vision Transformer.

Area of Science:

• Artificial Intelligence in Medicine
• Respiratory Medicine
• Signal Processing

Background:

• Respiratory diseases are a leading global cause of mortality, necessitating advanced diagnostic tools.
• Current methods for diagnosing respiratory conditions rely on specialized expertise and can be augmented by AI.
• Accurate identification of abnormal lung sounds is crucial for timely diagnosis and treatment of respiratory ailments.

Purpose of the Study:

• To develop and evaluate a novel artificial intelligence (AI) approach for the identification of abnormal respiratory sounds.
• To leverage audio-spectrograms and vision transformer models for enhanced respiratory sound classification.
• To compare the performance of the proposed method against existing state-of-the-art techniques.

Main Methods:

• Respiratory sounds were transformed into visual representations (spectrograms) using the short-time Fourier transform (STFT).
• An audio-spectrogram vision transformer (AS-ViT) model was employed to analyze these spectrograms for sound classification.
• The ICBHI 2017 database, comprising diverse lung sound recordings, was utilized for model training and validation.

Main Results:

• The AS-ViT model achieved high performance in respiratory sound detection across different data splits.
• Performance metrics included unweighted average recall and overall scores, demonstrating the model's effectiveness.
• The proposed AS-ViT method surpassed previous state-of-the-art results in classifying respiratory sounds.

Conclusions:

• The developed AS-ViT approach shows significant promise for accurate and automated detection of abnormal respiratory sounds.
• This AI-driven method can serve as a valuable tool to assist healthcare professionals in diagnosing lung disorders.
• Further research and validation could lead to widespread clinical application of AI in respiratory diagnostics.