Classification of Adventitious Sounds Combining Cochleogram and Vision Transformers

Area of Science:

• Medical Informatics
• Signal Processing
• Artificial Intelligence

Background:

• Early detection of respiratory irregularities is crucial for improving lung health and reducing mortality.
• Respiratory sound analysis is a key method for assessing lung condition and identifying abnormalities.

Purpose of the Study:

• To investigate the performance of the Vision Transformer (ViT) architecture using cochleograms as input for adventitious respiratory sound classification.
• To evaluate the novel combination of cochleogram input with ViT for respiratory sound analysis.

Main Methods:

• The study utilized the ICBHI dataset for evaluating the proposed methodology.
• Respiratory sounds were represented as cochleograms and fed into the ViT architecture.
• Classification performance was compared against state-of-the-art Convolutional Neural Network (CNN) approaches using various input features (spectrogram, MFCC, CQT, cochleogram).

Main Results:

• The combination of cochleogram input with the ViT architecture demonstrated superior classification performance compared to other methods.
• ViT, when applied to cochleograms, showed significant potential for reliable respiratory sound classification.
• The results confirm the effectiveness of this novel input-classifier pairing for adventitious sound classification.

Conclusions:

• The findings highlight the potential of ViT with cochleogram input for accurate and efficient respiratory sound classification.
• This research contributes to the development of intelligent automated techniques for augmenting the speed and effectiveness of respiratory disease detection.
• The study addresses a critical need for improved diagnostic tools in respiratory medicine.