Extubation Decision Support in Critical Care: A Multimodal Machine Learning Framework Integrating Segmented...

Area of Science:

• Critical Care Medicine
• Artificial Intelligence in Healthcare
• Medical Imaging Analysis

Background:

• Extubation failure in critically ill patients leads to poor outcomes.
• Current extubation decisions rely on subjective weaning tests and chest X-ray interpretations.
• There is a need for objective, data-driven decision support tools.

Purpose of the Study:

• To develop a feasible multimodal machine learning (ML) framework for extubation decision support.
• To integrate routinely available clinical data and chest X-ray images.
• To improve the accuracy of predicting extubation success.

Main Methods:

• A multimodal ML framework was developed using data from 921 patients.
• The framework integrated demographics, weaning measurements, radiographic assessments, and segmented chest X-rays.
• A stacking ensemble approach combined optimal base ML models for each data modality.

Main Results:

• The multimodal ensemble achieved 79.46% accuracy in predicting extubation outcomes.
• Weaning measurements, demographics, and radiographic assessments were key predictors.
• The ensemble model outperformed rule-based and single-modality models.

Conclusions:

• A multimodal ML framework using segmented chest X-rays and clinical data can support extubation decisions.
• This approach offers an objective, complementary tool without requiring extra measurements.
• Further prospective studies are necessary for validation.