Detection of Left Ventricular Systolic Dysfunction From Electrocardiographic Images

Area of Science:

• Cardiology
• Artificial Intelligence
• Medical Imaging

Background:

• Left ventricular (LV) systolic dysfunction significantly increases heart failure and mortality risks.
• Current ECG screening methods for LV dysfunction are limited by clinical availability.
• A novel deep learning approach for LV dysfunction screening using ECG images has been developed.

Purpose of the Study:

• To develop and validate a deep learning algorithm for detecting LV systolic dysfunction from ECG images.
• To assess the model's performance across diverse clinical settings and ECG formats.
• To identify ECG patterns indicative of LV systolic dysfunction.

Main Methods:

• A convolutional neural network was trained on 385,601 ECGs and echocardiograms.
• The model was validated internally and externally across multiple hospital systems and a prospective cohort.
• Gradient-weighted class activation mapping was used for signal localization.

Main Results:

• The model achieved high discrimination (AUROC 0.90-0.95) across various validation sets.
• ECG findings suggestive of LV dysfunction were associated with over 27-fold increased odds of the condition.
• Positive ECG screens predicted a 3.9-fold increased risk of future incident LV systolic dysfunction.

Conclusions:

• A deep learning model effectively identifies LV systolic dysfunction from ECG images.
• This method provides an automated and accessible screening strategy for LV systolic dysfunction.
• The approach holds particular promise for low-resource healthcare settings.