A personalized classification system for Holter registers

Area of Science:

• Biomedical Engineering
• Cardiology
• Data Science

Background:

• Long-term electrocardiogram (ECG) monitoring generates vast amounts of data, making manual analysis challenging.
• Identifying subtle patterns in Holter register data is crucial for diagnosing latent heart conditions.
• Automated analysis of ECG signals is needed to improve diagnostic efficiency and accuracy.

Purpose of the Study:

• To develop a personalized, long-term ECG classification framework for Holter registers.
• To automate the extraction and selection of representative 'master key-beats' from homogeneous ECG segments.
• To enhance the accuracy and efficiency of diagnosing heart diseases from extensive ECG data.

Main Methods:

• Developed a personalized ECG classification framework applicable to individual Holter registers.
• Implemented an automated system to extract 'master key-beats' from time frames of similar beats.
• Utilized exhaustive K-means clustering to determine the optimal number and selection of master key-beats.
• Validated the system on a benchmark database with cardiologist-labeled ECG beats.

Main Results:

• The automated system achieved over 99% average accuracy in classifying ECG beats.
• Results demonstrated consistency with manual labels provided by cardiologists.
• The framework proved efficient and robust in handling massive datasets and high-dimensional features.
• The selection of master key-beats was identified as critical for high-accuracy classification.

Conclusions:

• The proposed personalized ECG classification framework offers a robust and efficient solution for analyzing long-term Holter data.
• Automated extraction and selection of master key-beats significantly improve diagnostic accuracy.
• The system's high accuracy validates its effectiveness in managing large-scale ECG data for clinical applications.