Sputum deposition classification for mechanically ventilated patients using LSTM method based on airflow signals

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Medicine
• Respiratory Care

Background:

• Mechanically ventilated patients often require airway clearance techniques.
• Accurate and timely detection of sputum deposition is crucial for effective respiratory management.
• Current methods for monitoring sputum can be invasive or labor-intensive.

Purpose of the Study:

• To develop and validate a novel method for classifying sputum deposition in mechanically ventilated patients.
• To leverage wireless ventilation airflow signals and long-short-term memory (LSTM) networks for automated classification.
• To assess the performance and advantages of the proposed LSTM-based method compared to existing classifiers.

Main Methods:

• A wireless ventilation airflow signal collection system was designed and implemented.
• Two hundred sixty data groups from 15 intensive care unit patients were analyzed.
• A two-layer LSTM framework utilizing 11 airflow signal features was trained and cross-validated.
• Classification performance was evaluated using sensitivity, specificity, precision, accuracy, F1 score, and G score.

Main Results:

• The proposed LSTM method achieved an accuracy of 84.7 ± 4.1% in classifying sputum deposition.
• The LSTM method demonstrated superior performance compared to logistic regression, random forest, naive Bayes, support vector machine, and K-nearest neighbor classifiers.
• Ventilation airflow signal analysis offers convenience and low complexity for sputum classification.

Conclusions:

• The novel LSTM-based method provides an accurate and efficient approach for sputum deposition classification in mechanically ventilated patients.
• The system's ability to integrate with intelligent devices facilitates timely alerts for medical staff, reducing workload.
• This technology holds significant potential for enhancing automation and efficiency in respiratory care, particularly relevant during pandemics like COVID-19.