A Study of R-R Interval Transition Matrix Features for Machine Learning Algorithms in AFib Detection

Area of Science:

• Cardiology
• Biomedical Engineering
• Machine Learning

Background:

• Atrial Fibrillation (AFib) presents diagnostic challenges due to subtle, unpredictable symptoms.
• Current diagnostic methods can be delayed, necessitating advanced prediction systems.
• R-R interval variability is a key indicator for AFib detection.

Purpose of the Study:

• To develop a predictive system for Atrial Fibrillation using R-R interval variability.
• To introduce and evaluate the transition matrix as a novel feature for R-R variability analysis.
• To systematically analyze feature significance using multiple segmentation schemes and importance measures.

Main Methods:

• Utilized the MIT-BIH dataset, segmented into 5-s, 10-s, and 25-s subsets.
• Extracted 21 features, including novel transition matrix features.
• Trained and evaluated 11 machine learning classifiers, employing permutation and tree-based feature importance with Leave-One-Person-Out Cross Validation.

Main Results:

• Classifiers using the 25-s segmentation scheme achieved the highest accuracies, with Gradient Boosting models exceeding 96%.
• Gradient Boosting and Random Forest models demonstrated superior performance across all segmentation schemes.
• Transition matrix features were identified as highly significant, particularly with longer data segments.

Conclusions:

• The 25-s segmentation scheme combined with gradient boosting models offers a highly accurate approach for AFib prediction.
• The transition matrix is a valuable feature for quantifying R-R interval variability in AFib detection.
• This research provides a robust framework for improving early and accurate diagnosis of Atrial Fibrillation.