Prediction of weaning from mechanical ventilation using Convolutional Neural Networks

Area of Science:

• Critical Care Medicine
• Artificial Intelligence in Healthcare
• Machine Learning for Clinical Decision Support

Background:

• Mechanical ventilation weaning is crucial for critically ill patients in Intensive Care Units (ICUs).
• Prolonged mechanical ventilation or premature extubation increases patient risks and healthcare costs.
• Developing accurate methods to predict extubation readiness is essential for optimizing patient care.

Purpose of the Study:

• To develop a decision support model for predicting extubation readiness.
• To utilize routinely recorded patient data for accurate predictions.
• To enhance clinical decision-making regarding mechanical ventilation liberation.

Main Methods:

• Deployment of Convolutional Neural Networks (CNN) for treatment action prediction.
• Utilizing historical Intensive Care Unit (ICU) data from the MIMIC-III database.
• Performing feature importance analysis using the DeepLIFT method and implementing counterfactual explanations.

Main Results:

• The CNN model achieved 86% accuracy and an AUC-ROC of 0.94 in predicting extubation readiness.
• Feature importance analysis confirmed the model uses clinically meaningful predictors.
• Counterfactual explanations provide insights into feature changes for desired outcomes.

Conclusions:

• The developed CNN model demonstrates high accuracy in predicting extubation readiness.
• The model's reliance on clinically relevant features supports its utility.
• AI-driven decision support can aid clinicians in optimizing the weaning process and patient outcomes.