Prospective validation of a deep learning electrocardiogram algorithm for the detection of left ventricular systolic...

Area of Science:

• Cardiology
• Artificial Intelligence
• Medical Diagnostics

Background:

• Undetected left ventricular (LV) dysfunction is common in patients undergoing routine electrocardiograms (ECG).
• Identifying patients with LV dysfunction can be challenging, necessitating advanced diagnostic methods.

Purpose of the Study:

• To validate a deep learning algorithm for predicting low ejection fraction (EF ≤ 35%) from 12-lead ECGs.
• To assess the algorithm's performance in a large prospective cohort and its potential for identifying patients with LV dysfunction.

Main Methods:

• The deep learning algorithm was applied to ECGs from the Mayo Clinic ECG laboratory.
• Algorithm performance was evaluated in patients with and without recent echocardiographic assessments.
• Sensitivity, specificity, and accuracy were calculated for predicting EF ≤ 35%.

Main Results:

• The algorithm demonstrated high accuracy (86.5%), sensitivity (82.5%), and specificity (86.8%) in detecting EF ≤ 35%.
• Among patients with no prior echocardiogram, the algorithm identified 3.5% with suspected low EF.
• Exploratory analysis suggested NT-pro-BNP assessment could reduce false positives.

Conclusions:

• A deep learning algorithm effectively detects depressed LV function using routine ECGs in clinical practice.
• Further validation is needed in diverse patient groups and for assessing clinical and economic impact.
• The algorithm shows promise for improving the identification of patients with potential cardiac dysfunction.