Classification of atrial fibrillation episodes from sparse electrocardiogram data.

Area of Science:

• Cardiology
• Biomedical Engineering
• Signal Processing

Background:

• Atrial fibrillation (AF) is the most prevalent cardiac arrhythmia.
• Detecting spontaneous AF termination is crucial for patient management.
• Sparse data presents challenges in accurately classifying AF episodes.

Purpose of the Study:

• To apply the Classification and Regression Tree (CART) technique for AF termination detection.
• To develop an efficient method for classifying AF episodes using sparse data.
• To evaluate the accuracy of CART in distinguishing between different AF termination patterns.

Main Methods:

• Utilized Electrocardiogram (ECG) recordings from the PhysioNet database.
• Employed continuous wavelet transform for compact ECG signal representation.
• Extracted key ECG quantifiers and applied Principal Component Analysis (PCA).
• Classified AF episodes into Nonterminating (N), Soon terminating (S), and Terminating (T) using CART.

Main Results:

• Continuous wavelet transform achieved a compact ECG representation, reducing data dimensionality.
• Efficient algorithms were developed for beat detection and feature extraction.
• Principal Component Analysis (PCA) facilitated the creation of a concise feature set.
• Achieved classification accuracies exceeding 90% for AF episodes.

Conclusions:

• A customized wavelet representation significantly reduced signal entropy.
• The CART technique demonstrated high accuracy in classifying N vs T and S vs T AF episodes.
• CART is a viable and accurate method for analyzing sparse ECG data in AF detection.