AI in Pediatrics

Windreich Department of AI & Human Health, Instructor

2025-Present

Problem

AI models in healthcare are predominantly developed in adult populations and often fail to generalize to neonates and children, where physiology is rapidly evolving, disease presentation is heterogeneous, and labeled datasets are limited.

There is a critical need for age-specific AI approaches that can support early diagnosis, risk stratification, and outcome prediction in pediatric care — particularly in high-acuity neonatal settings.

Approach

In collaboration with neonatology and cardiology clinicians, I develop AI models tailored to pediatric and neonatal datasets, focusing on clinically meaningful prediction tasks.

Current areas of work include:

• Transient Tachypnea of the Newborn (TTN): Developing models using chest X-ray imaging and clinical variables to predict outcomes. We identified poor translation of adult-based X-ray models for neonatal application.
• Echocardiogram-based cardiac assessment: Applying machine learning methods to echocardiographic image-waveforms to support detection of cardiac abnormalities.
• Neonatal apnea detection: Developed GUI to label ground truth apneas status using high sampling rate respiratory waveforms recorded from NICU. This will facilitate use of continuous physiological monitoring data to identify early signs of respiratory instability.

These projects emphasize careful pre-processing, multimodal data integration, and age-based data stratification appropriate for small and specialized datasets.

Impact

By designing AI systems specifically for neonatal and pediatric populations, this work aims to improve early risk identification and clinical decision support in settings where timely intervention is critical.

These efforts contribute toward building scalable, data-driven tools that enhance precision monitoring.