Electrocardiographic Classification using Deep Learning with Lead Switching.

Area of Science:

• Cardiology
• Biomedical Engineering
• Artificial Intelligence

Background:

• Electrocardiogram (ECG) signal analysis is crucial for diagnosing cardiac conditions.
• Current ECG classification methods often rely on fixed lead configurations, potentially limiting diagnostic performance.
• Abnormalities in ECG rhythm and morphology require accurate classification for effective patient management.

Purpose of the Study:

• To develop and evaluate a novel neural network-based method for ECG classification using a dynamic lead-switching strategy.
• To improve the accuracy of identifying cardiac abnormalities from single-lead ECG signals.
• To determine the optimal sequence of leads for enhanced classification performance.

Main Methods:

• A neural network model was designed to process sequences of single-lead ECG signals.
• The model learns to classify ECGs by adaptively selecting the most informative lead at each step.
• An optimal lead order was determined using the area under the receiver-operating characteristic curve (AUC) metric.

Main Results:

• The proposed lead-switching method demonstrated significantly improved AUCs compared to fixed single-lead ECG analysis.
• The approach achieved classification performance comparable to traditional 12-lead ECGs for several diagnoses.
• Experiments were conducted on a dataset of 6,877 ECG recordings covering nine distinct diagnoses.

Conclusions:

• Dynamic lead switching in ECG analysis offers a promising approach to enhance classification accuracy.
• This method provides a flexible and potentially more efficient alternative to fixed multi-lead ECG systems.
• The findings suggest that adaptive lead observation can improve the detection of cardiac abnormalities.