Abstract
Critical congenital heart disease (CCHD) is a leading cause of neonatal mortality, yet current screening protocols often fail to detect certain life-threatening defects. This data collection and waveform analysis system is relatively inexpensive and can be used to enhance future CCHD screening algorithms by synchronizing high-fidelity pulse oximetry data from two anatomical sites.
Introduction
Critical congenital heart disease (CCHD) affects approximately 0.8% of all births. While universal screening using oxygen saturation (SpO2) has saved many lives, current methods still miss nearly 900 infants annually in the United States. A significant blind spot exists for conditions like Coarctation of the Aorta (CoA), where a newborn may maintain normal oxygen levels despite severely compromised blood flow (perfusion) to the lower body.
Standard clinical monitors typically provide only de-identified summary data, losing the rich physiological information contained in raw photoplethysmography (PPG) waveforms. To improve detection, researchers require synchronized, high-fidelity data from both the right hand (pre-ductal) and a foot (post-ductal).
System Design and Methodology
Researchers developed a custom hardware and software architecture to bridge this data gap. The system utilizes two Nonin WristOx2 3150 pulse oximeters paired via Bluetooth with a "central aggregator"—a Pi-top™ laptop powered by a Raspberry Pi microcomputer.
A custom Linux-based environment automates data collection, ensures accurate time-stamping between the independent sensors, and provides real-time visualization for clinicians. The workflow is designed for ease of use by non-technical staff: a technician attaches the probes, enters a study ID and the patient's inspired oxygen (FiO2), and the system handles high-resolution recording and storage.
Multi-Center Deployment and Results
The system was deployed across five hospitals, collecting data from 190 neonates, including 130 healthy infants and 60 with suspected congenital heart disease. In total, 1665 minutes of high-fidelity data were recorded, capturing 31 confirmed CCHD cases.
By capturing raw PPG waveforms, the system allows for the calculation of advanced physiological markers:
- Radiofemoral Delay: The time lag between the pulse reaching the hand versus the foot, a vital marker for obstructions like CoA.
- Waveform Morphology: High-resolution data revealed features like the "dicrotic notch," which changed significantly as heart defects progressed—detail that is lost in standard clinical readouts.
- Dynamic Perfusion Tracking: In one case study, the system tracked a dramatic drop in the foot's Perfusion Index (PAI) from 96 to 36 as an infant's condition evolved.
Discussion and Future Directions
The use of off-the-shelf components like the Raspberry Pi makes the system inexpensive and portable, suitable for diverse clinical settings. While technical hurdles like Bluetooth jitter and manual data transfer steps remain, the system successfully extracted nine distinct features from the raw data, including heart rate variability and pulse amplitude indices.
The ultimate goal is to use this high-fidelity dataset to train machine learning models. These models could eventually provide real-time, automated CCHD screening that is far more sensitive than current SpO2-only protocols, potentially identifying hundreds of cases that are currently missed.
Summary updated