Novel heuristic search for ventricular arrhythmia detection using normalized cut clustering.

Area of Science:

• Cardiology
• Biomedical Engineering
• Machine Learning

Background:

• Accurate arrhythmia detection from long-term ECG Holter recordings is challenging due to large data volumes and imbalanced class distributions.
• Existing heartbeat classification methods struggle to effectively address these issues.
• Ventricular cardiac arrhythmias require precise identification for effective patient management.

Purpose of the Study:

• To introduce a novel heuristic-search-based clustering method for discriminating ventricular cardiac arrhythmias.
• To address the limitations of existing methods in handling large datasets and imbalanced classes in ECG analysis.
• To provide an unsupervised approach for heartbeat classification.

Main Methods:

• A heuristic-search-based clustering algorithm utilizing the normalized cut criterion was developed.
• The method iteratively groups nodes based on a maximum similarity value.
• Initial algorithm parameters were set using a kernel density estimator for unsupervised operation.
• The MIT/BIH arrhythmia database was used for performance evaluation.

Main Results:

• The proposed heuristic-search clustering demonstrated adequate performance in classifying ventricular arrhythmias.
• The method proved effective even when dealing with highly unbalanced classes within the dataset.
• Heartbeat labeling was achieved through analysis of the MIT/BIH arrhythmia database.

Conclusions:

• The developed unsupervised heuristic-search clustering offers a viable solution for heartbeat classification in ECG Holter recordings.
• This approach shows promise for improving the accuracy of arrhythmia detection, particularly in challenging scenarios with imbalanced data.
• Further research can explore optimizations and applications of this method in clinical settings.