Lead-Specific Performance for Atrial Fibrillation Detection in Convolutional Neural Network Models Using Sinus Rhythm...

Area of Science:

• Cardiology
• Artificial Intelligence in Medicine
• Medical Diagnostics

Background:

• Atrial fibrillation (AF) detection is crucial for preventing stroke.
• Convolutional neural network (CNN) models show promise for AF detection using electrocardiograms (ECGs).
• The optimal ECG lead configuration for CNN-based AF detection using sinus rhythm ECG (SR-ECG) remains to be determined.

Purpose of the Study:

• To evaluate the diagnostic performance of a CNN model for AF detection using various SR-ECG lead configurations.
• To compare the effectiveness of 8-lead, single-lead, and double-lead SR-ECGs for CNN-based AF detection.
• To identify potential alternative ECG lead combinations for efficient AF screening.

Main Methods:

• A retrospective analysis of a prospective cohort study involving 19,170 new patients.
• Development and 5-fold cross-validation of a CNN model using SR-ECGs from 616 AF cases and 3,412 SR cases.
• Evaluation of CNN model performance using 8-lead, single-lead, and double-lead (specifically leads I and V1) SR-ECGs.

Main Results:

• The CNN model achieved an Area Under the Curve (AUC) of 0.872 for the 8-lead ECG.
• A double-lead ECG using leads I and V1 yielded a comparable AUC of 0.871.
• The odds ratio for AF detection was 15.24 for the 8-lead ECG and 14.34 for the double-lead (I, V1) ECG.

Conclusions:

• A CNN model can effectively detect AF using SR-ECG data.
• A double-lead ECG (leads I and V1) provides diagnostic performance comparable to an 8-lead ECG for AF detection.
• The double-lead ECG (I, V1) presents a potential alternative for AF screening, simplifying data acquisition.