Detection of Right and Left Ventricular Dysfunction in Pediatric Patients Using Artificial Intelligence-Enabled ECGs.

Area of Science:

• Pediatric cardiology
• Artificial intelligence in medicine
• Electrocardiography

Background:

• Early detection of left and right ventricular systolic dysfunction (LVSD and RVSD) in children is crucial for reducing morbidity and mortality.
• Existing AI algorithms for LVSD/RVSD detection are validated in adults, but their efficacy in pediatric populations is uncertain.

Purpose of the Study:

• To develop novel artificial intelligence-enabled ECG algorithms for detecting LVSD and RVSD in pediatric patients.
• To evaluate the performance of these pediatric-specific models compared to adult-derived models.

Main Methods:

• Utilized a dataset of 10,142 pediatric patients (age ≤18) with 12-lead ECGs and transthoracic echocardiograms.
• Developed novel AI models for detecting LVSD (defined by LVEF thresholds) and RVSD.
• Compared performance metrics (AUC, sensitivity, specificity, PPV, NPV) of pediatric models against adult-derived models.

Main Results:

• Novel pediatric models achieved excellent test areas under the curve (AUC): 0.93 for LVEF ≤35% and 0.88 for LVEF <50%.
• The pediatric model for LVEF <50% demonstrated high sensitivity (0.85) and specificity (0.80), with a negative predictive value of 0.995.
• A pediatric model for RVSD detection reached an AUC of 0.90.
• Adult-derived models showed good performance (AUC 0.87 for LVEF ≤35%) but were outperformed by pediatric-specific models.

Conclusions:

• AI-enabled ECG analysis accurately detects both LVSD and RVSD in pediatric patients.
• Pediatric-specific AI models offer superior performance compared to adult-trained models for diagnosing these conditions in children.
• These findings support the clinical utility of AI-ECG for early identification of pediatric heart dysfunction.