Multiple high-regional-incidence cardiac disease diagnosis with deep learning and its potential to elevate...

Area of Science:

• Cardiology
• Artificial Intelligence
• Medical Informatics

Background:

• AI-aided cardiac disease diagnosis is limited by the lack of large-scale datasets.
• The potential of AI-based ECG diagnosis to assist cardiologists requires further investigation.

Purpose of the Study:

• To construct a comprehensive dataset of 12-lead ECGs for common cardiac conditions.
• To develop and evaluate a deep learning model for AI-aided diagnosis of multiple cardiac diseases.
• To assess the impact of AI assistance on cardiologist performance.

Main Methods:

• A large-scale dataset of 162,622 12-lead ECGs was created, spanning January 2018 to March 2021.
• A deep learning model was developed for the clinical ECG diagnosis of 15 cardiac abnormalities.
• Model performance was evaluated against board-certified cardiologists on a re-annotated dataset.

Main Results:

• The deep learning model achieved 88.216% accuracy and an average AUC ROC score of 0.961 for diagnosing 15 cardiac abnormalities.
• The AI model's performance exceeded that of cardiologists in a non-reference group.
• Cardiologist accuracy and efficiency improved by 13.5% and 69.9% respectively when aided by the AI model's labels.

Conclusions:

• The developed deep learning model offers a viable solution for AI-aided diagnosis of cardiac diseases.
• AI assistance can significantly enhance both the accuracy and efficiency of clinical cardiologists.
• This approach has practical applications for improving cardiac disease diagnosis systems.