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

A novel sputum deposition classification method for mechanically ventilated patients based on the long-short-term memory network (LSTM) method was proposed in this study. A wireless ventilation airflow signals collection system was designed and used in this study. The ventilation airflow signals were collected wirelessly and used for sputum deposition classification. Two hundred sixty data groups from 15 patients in the intensive care unit were compiled and analyzed. A two-layer LSTM framework and 11 features extracted from the airflow signals were used for the model training. The cross-validations were adopted to test the classification performance. The sensitivity, specificity, precision, accuracy, F1 score, and G score were calculated. The proposed method has an accuracy of 84.7 ± 4.1% for sputum and non-sputum deposition classification. Moreover, compared with other classifiers (logistic regression, random forest, naive Bayes, support vector machine, and K-nearest neighbor), the proposed LSTM method is superior. In addition, the other advantages of using ventilation airflow signals for classification are its convenience and low complexity. Intelligent devices such as phones, laptops, or ventilators can be used for data processing and reminding medical staff to perform sputum suction. The proposed method could significantly reduce the workload of medical staff and increase the automation and efficiency of medical care, especially during the COVID-19 pandemic.