Towards End-to-End ECG Classification With Raw Signal Extraction and Deep Neural Networks.

Area of Science:

• Cardiology
• Biomedical Engineering
• Artificial Intelligence

Background:

• Electrocardiogram (ECG) signal analysis is crucial for diagnosing heart conditions.
• Current methods often rely on handcrafted features extracted by medical experts, which can be time-consuming and suboptimal.
• Automated and accurate heartbeat classification remains a significant challenge in clinical practice.

Purpose of the Study:

• To propose a deep learning framework for end-to-end classification of raw ECG signals into different heartbeat types.
• To develop a method that automatically extracts features and classifies heartbeats, eliminating the need for manual feature engineering.
• To achieve patient-independent ECG classification with high accuracy.

Main Methods:

• Utilizing deep neural networks for automated feature extraction and classification of ECG signals.
• Implementing a signal alignment strategy to process raw time-domain sample points.
• Extracting consecutive vectors representing complete heartbeat cycles, including P, QRS, and T waves.

Main Results:

• The proposed patient-independent classifier demonstrated at least 10% higher sensitivity in detecting supraventricular and ventricular ectopic beats compared to state-of-the-art methods at the same specificity.
• Achieved superior sensitivity and specificity across a wide range of operating points compared to existing classifiers.
• The deep learning approach provided performance comparable to patient-specific classifiers while maintaining patient independence.

Conclusions:

• The developed deep learning method offers an effective end-to-end solution for ECG heartbeat classification.
• Signal alignment and deep neural networks optimize ECG representation for enhanced diagnostic accuracy.
• This approach presents a promising, accurate, and patient-independent alternative for arrhythmia detection.