Development and Validation of a Class Imbalance-Resilient Cardiac Arrest Prediction Framework Incorporating...

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Healthcare
• Time-Series Data Analysis

Background:

• Existing AI models for cardiac arrest (CA) prediction struggle with data imbalance, efficiency, accuracy, and explainability.
• Multiscale feature extraction, while used in other AI applications, has not been applied to diagnostic CA prediction.

Purpose of the Study:

• To present a novel framework for CA prediction that addresses the limitations of current AI models.
• To utilize multiscale feature aggregation via Independent Component Analysis (ICA) for improved diagnostic performance.

Main Methods:

• The Pareto optimal StrataChron Pyramid Fusion Framework (SPFF) was developed for multi-scale temporal feature aggregation.
• Independent Component Analysis (ICA) was integrated to reduce data redundancy and enhance efficiency.
• The framework was validated using the public MIMIC IV dataset.

Main Results:

• The proposed model demonstrated resilience to data imbalance.
• Explainability was enhanced using SHapley Additive exPlanations (SHAP), Partial Dependence Plots (PDP), and Individual Conditional Expectation (ICE).
• The multilayer perceptron (MLP) achieved high performance metrics: accuracy (0.982), precision (0.969), recall (0.989), F1-score (0.979), and AUROC (0.998).

Conclusions:

• The developed method effectively predicts CA across various time windows.
• It offers a robust solution to the challenges of efficiency, accuracy, and explainability in CA prediction models.
• The approach captures both short- and long-term patient data dependencies, potentially improving clinical decision-making and patient outcomes.