Artificial intelligence for ventricular arrhythmia capability using ambulatory electrocardiograms.

Area of Science:

• Cardiology
• Artificial Intelligence
• Medical Diagnostics

Background:

• Current clinical guidelines for implantable cardioverter defibrillators (ICDs) have limitations in accurately stratifying ventricular arrhythmia (VA) risk.
• This inaccuracy leads to significant patient morbidity and mortality.
• Artificial intelligence (AI) presents a novel approach for VA risk stratification using electrocardiograms (ECGs).

Purpose of the Study:

• To develop and validate a deep neural network (DNN) for VA risk stratification.
• To assess the capability of AI to determine VA risk from routine ambulatory ECGs.

Main Methods:

• A multicentre case-control study utilized an open-source ResNet-50-based DNN, VA-ResNet-50.
• The model analyzed three-lead, 24-hour ambulatory ECGs to predict VA capability.
• 270 adult patients (159 with VA) were included, with ECGs collected up to 1.6 years prior to VA events.

Main Results:

• VA-ResNet-50 achieved an accuracy of 0.76 and an F1 score of 0.79 in classifying VA capability from ECGs.
• The model demonstrated an area under the receiver operator curve of 0.8.
• Individuals identified as high-risk by the AI had a 2.87 times higher relative risk of VA.

Conclusions:

• Ambulatory ECGs contain valuable risk signals for VA stratification when analyzed by VA-ResNet-50.
• The AI model's performance surpasses current medical guidelines.
• This AI-driven approach holds promise for optimizing the allocation of life-saving ICDs.