The Effectiveness of a Deep Learning Model to Detect Left Ventricular Systolic Dysfunction from Electrocardiograms

Area of Science:

• Cardiology
• Artificial Intelligence
• Medical Imaging

Background:

• Left ventricular (LV) dysfunction is a critical indicator of cardiac health.
• Early and accurate detection of LV dysfunction is crucial for timely intervention.
• Electrocardiograms (ECGs) are widely available but traditionally limited in diagnosing LV dysfunction.

Purpose of the Study:

• To develop and validate a deep learning model for detecting LV dysfunction from ECGs.
• To assess the impact of AI-assisted interpretation on cardiologists' diagnostic accuracy.
• To investigate the specific ECG features utilized by the AI model.

Main Methods:

• A convolutional neural network (CNN) was trained on 23,801 ECGs from patients with known LV dysfunction (ejection fraction < 40%).
• The model was validated on an independent test set of 7,196 ECGs.
• The diagnostic performance of 7 cardiologists was compared with and without AI model support.

Main Results:

• The deep learning model achieved a high diagnostic accuracy, with an area under the receiver operating characteristic curve of 0.945.
• Cardiologists' accuracy in predicting LV dysfunction improved from 78.0% to 88.0% when assisted by the AI model (P = 0.02).
• Sensitivity analysis indicated the model primarily focused on the QRS complex for LV dysfunction detection.

Conclusions:

• A deep learning model can effectively detect LV dysfunction from ECGs with high accuracy.
• AI-powered decision support significantly enhances cardiologists' ability to diagnose LV dysfunction using ECGs.
• This AI tool holds promise for improving the non-invasive diagnosis of cardiac conditions.