Related Experiment Video
Updated: Oct 10, 2025
07:34
Fetal Mouse Cardiovascular Imaging Using a High-frequency Ultrasound 30/45MHZ System
Published on: May 5, 2018
11.8K
Enhanced Critical Congenital Cardiac Disease Screening by Combining Interpretable Machine Learning Algorithms
Summary
Newborn screening for Critical Congenital Heart Disease (CCHD) can be improved. Adding machine learning analysis of pulse oximetry data enhances detection rates for CCHD, helping more infants receive timely care.
Area of Science:
- Biomedical Engineering
- Neonatal Medicine
- Artificial Intelligence in Healthcare
Background:
- Current Critical Congenital Heart Disease (CCHD) screening using only oxygen saturation (SpO2) misses approximately 900 US newborns annually.
- CCHD cases with systemic blood flow obstruction, like Coarctation of the Aorta (CoA), are particularly challenging for SpO2-only screening.
Purpose of the Study:
- To investigate the utility of additional pulse oximetry features beyond SpO2 for improved CCHD detection.
- To develop and validate a machine learning (ML) model integrating multiple pulse oximetry parameters for enhanced CCHD screening.
Main Methods:
- Employed Recursive Feature Elimination (RFE) with interpretable ML algorithms to identify optimal features for CCHD detection.
- Integrated the trained ML model with the existing SpO2-based screening algorithm to create an enhanced system.
Main Results:
- The enhanced CCHD screening system demonstrated a significant improvement in sensitivity, increasing it by approximately 10 percentage points.
- The enhanced system maintained minimal to no negative impact on specificity compared to the SpO2-alone method.
Conclusions:
- This study provides proof of concept for an ML-driven approach to CCHD screening that combines multiple pulse oximetry features.
- The proposed method offers a promising strategy to improve CCHD detection rates in newborns without substantially increasing false positives.
