Interpretation of EKG with Image Recognition and Convolutional Neural Networks.

Area of Science:

• Cardiology
• Artificial Intelligence
• Medical Imaging

Background:

• Electrocardiograms (EKG) are crucial for cardiovascular diagnosis and management.
• Current automated EKG interpretation algorithms often lack generalizability due to machine-specific proprietary signals.
• There is a need for universal and accurate automated EKG analysis tools.

Purpose of the Study:

• To develop a generalizable image recognition model for interpreting standard 12-lead electrocardiogram (EKG) strips.
• To apply a convolutional neural network (CNN) for classifying EKGs into clinically significant diagnostic categories.
• To overcome the limitations of proprietary, machine-specific EKG interpretation algorithms.

Main Methods:

• Development of a convolutional neural network (CNN) model based on a MobileNetV3 architecture.
• Training the CNN from scratch using a publicly available, labeled dataset of 12-lead EKGs.
• Classifying EKGs into seven distinct, clinically relevant diagnostic categories.

Main Results:

• The developed image recognition model achieved varying precision per class, ranging from 52% to 91%.
• The model demonstrated a novel approach to EKG interpretation by treating it as an image recognition problem.
• Successful classification of 12-lead EKGs into seven diagnostic classes was achieved.

Conclusions:

• An image recognition model, specifically a CNN, can be effectively trained to interpret standard EKG strips.
• This approach offers a potentially more generalizable alternative to existing machine-specific EKG interpretation algorithms.
• The study highlights the potential of artificial intelligence in advancing cardiovascular diagnostics through novel EKG analysis methods.