Convolutional Neural Network for Breathing Phase Detection in Lung Sounds.

Area of Science:

• Medical technology
• Artificial intelligence in healthcare
• Respiratory medicine

Background:

• Accurate detection of breathing phases in lung sound recordings is crucial for diagnosing respiratory conditions.
• Current methods for breathing phase detection can be subjective and time-consuming.
• Deep learning offers a potential solution for automated and objective analysis of lung sounds.

Purpose of the Study:

• To develop and validate a deep learning algorithm for automated breathing phase detection in lung sound recordings.
• To compare the algorithm's performance against manual annotations by experienced researchers.
• To assess the algorithm's accuracy, sensitivity, and specificity in identifying inspiratory and expiratory phases.

Main Methods:

• A convolutional neural network (CNN) model was developed using spectrograms as input features.
• The algorithm was trained and evaluated on three large, novel datasets of lung sound recordings.
• Performance was assessed using discrete agreement counts and time-based agreement metrics (pseudo-kappa values).

Main Results:

• The algorithm demonstrated high agreement with human experts: 97% for inspiration and 87% for expiration.
• Time-based analysis yielded pseudo-kappa values ranging from 0.63 to 0.88.
• The algorithm achieved an average sensitivity of 97% and an average specificity of 84%.

Conclusions:

• The developed deep learning algorithm achieves human-level performance in detecting breathing phases from lung sound recordings.
• The algorithm is a valid and accurate tool for automated respiratory sound analysis.
• This technology has the potential to improve the efficiency and objectivity of respiratory diagnostics.