Pediatric sex estimation using AI-enabled ECG analysis: influence of pubertal development

Area of Science:

• Artificial Intelligence in Medicine
• Pediatric Cardiology
• Biomedical Signal Processing

Background:

• Artificial intelligence (AI) has demonstrated efficacy in predicting patient sex from electrocardiograms (ECGs) in adults.
• The correlation between ECG characteristics and sex hormones suggests potential applications in pediatric populations.
• Understanding the influence of pubertal development on ECG-sex prediction is crucial for pediatric applications.

Purpose of the Study:

• To evaluate the capability of AI-enabled ECGs to predict sex in pediatric patients.
• To investigate the impact of pubertal development stages on the accuracy of AI-based sex prediction from ECGs.
• To compare the performance of a de novo pediatric AI model with a transfer learning model derived from adult data.

Main Methods:

• Development of AI models using convolutional neural networks trained on 10-second, 12-lead pediatric ECGs.
• Training and validation datasets were established from 90,133 unique pediatric patient ECGs (age ≤18 years).
• Subgroup analysis was conducted across prepubertal (0-7), peripubertal (8-14), and postpubertal (15-18) age groups.

Main Results:

• The de novo pediatric AI model achieved 81% accuracy and an Area Under the Curve (AUC) of 0.91 across the entire test cohort.
• Model performance varied by pubertal stage, with higher discriminatory ability in postpubertal (AUC=0.98) and peripubertal (AUC=0.91) children compared to prepubertal children (AUC=0.67).
• No significant performance difference was found between the de novo and transfer learning models.

Conclusions:

• AI-enabled ECG interpretation can accurately estimate sex in peripubertal and postpubertal children.
• The accuracy of AI-based sex prediction from ECGs is influenced by the stage of pubertal development.
• AI holds promise for non-invasive sex estimation in pediatric care, particularly in older children and adolescents.