Identifying patients with acute aortic dissection using an electrocardiogram with convolutional neural network

Area of Science:

• Cardiology
• Artificial Intelligence
• Medical Diagnostics

Background:

• Electrocardiography (ECG) combined with artificial intelligence (AI) offers potential for initial screening of aortic dissection (AD).
• Achieving a high positive predictive rate (PPR) for AI-based AD detection remains a significant challenge.

Purpose of the Study:

• To evaluate the performance of a convolutional neural network (CNN) model for detecting aortic dissection (AD) using digital 12-lead ECG data.
• To assess the impact of different ECG lead configurations and patient risk factors on the CNN model's diagnostic accuracy and PPR.

Main Methods:

• A retrospective analysis of a prospective cohort study (Shinken Database, 2010-2017, N=19,170) utilizing digital 12-lead ECGs.
• A CNN model was trained and validated using five-fold cross-validation on eight-lead, single-lead, and double-lead ECG configurations for AD detection.
• Performance metrics including area under the curve (AUC) and positive predictive rate (PPR) were assessed across the entire cohort and in subgroups with elevated D-dimer and hypertension history.

Main Results:

• The CNN model achieved an AUC of 0.936 with eight-lead ECGs for AD detection.
• In the overall cohort, the model had a 7% PPR and 86% sensitivity for AD.
• Applying the CNN to patients with D-dimer levels ≥1 μg/dL and hypertension history increased PPR to 35% (86% sensitivity).
• A single V1 lead ECG configuration demonstrated high diagnostic performance (AUC: 0.933) and improved PPR to 38% in the same high-risk subgroup.

Conclusions:

• The developed CNN model demonstrates effective AD detection using ECG data, achieving over 30% PPR in high-risk patients while maintaining sensitivity.
• A single V1 lead ECG offers a simplified yet highly effective approach for AI-driven AD screening within the CNN framework.
• These findings suggest that AI-powered ECG analysis can significantly enhance the initial screening of aortic dissection, particularly in at-risk populations.