A Comparative Study on Predication of Appropriate Mechanical Ventilation Mode through Machine Learning Approach

Area of Science:

• Critical Care Medicine
• Biomedical Engineering
• Data Science

Background:

• Mechanical ventilation requires precise setting selection by critical care therapists.
• Ventilation mode selection is patient-specific and requires continuous interaction.
• Optimizing ventilation modes is crucial for effective patient management in intensive care units.

Purpose of the Study:

• To outline ventilation mode settings and identify the best machine learning (ML) method for predicting optimal modes.
• To develop a deployable ML model for per-breath ventilation mode selection.
• To compare the performance of various ML algorithms for this task.

Main Methods:

• Utilized per-breath patient data, including inspiratory/expiratory tidal volume, minimum pressure, and PEEP.
• Preprocessed data into a feature set for ML model training and testing (30% test size).
• Trained and evaluated six ML algorithms based on accuracy, F1 score, sensitivity, and precision.

Main Results:

• The Random-Forest Algorithm achieved the highest accuracy and precision in predicting ventilation modes.
• All tested ML algorithms were compared, with Random-Forest outperforming others.
• The study identified Random-Forest as a suitable technique for predicting optimal ventilation modes.

Conclusions:

• The Random-Forest algorithm is effective for predicting optimal mechanical ventilation modes on a per-breath basis.
• Accurate ML models require training with relevant, high-quality patient data.
• Future applications of ML, including deep learning, could optimize other mechanical ventilation parameters.