[A lung sound classification model with a spatial and channel reconstruction convolutional module]

Area of Science:

• Respiratory acoustics
• Biomedical signal processing
• Machine learning in healthcare

Context:

• Lung sound analysis is crucial for diagnosing respiratory conditions.
• Accurate classification of lung sounds aids in early disease detection.
• Existing methods may struggle with complex sound patterns.

Purpose:

• To develop a novel convolutional neural network (CNN) model for precise lung sound classification.
• To integrate a spatial-channel reconstruction convolution (SCConv) module for enhanced feature extraction.
• To evaluate the model's performance on the ICBHI2017 dataset.

Summary:

• A CNN architecture incorporating SCConv was proposed for lung sound analysis.
• A feature extraction method combining dual tunable Q-factor wavelet transform (DTQWT) and triple Wigner-Ville transform (WVT) was employed.
• The model was tested for classifying normal, crackles, wheezes, and crackles with wheezes.

Impact:

• The proposed model achieved high accuracy (85.68%), sensitivity (93.55%), specificity (86.79%), and F1 score (90.51%).
• Demonstrates significant potential for improving the accuracy of automated respiratory diagnostics.
• Effective in distinguishing between normal and abnormal lung sounds, aiding clinical decision-making.