Project Details

Computer Vision AI for Neuromonitoring in Neonatal Intensive Care

Felix Richter, MD, PhD

Summary

BACKGROUND: Currently, neurologic changes such as neonatal encephalopathy, pain control, and sedation are primarily assessed by physical examination. However, this approach is deeply flawed since physical examination is performed at set time intervals, is highly subjective, and may not discern subtle or subacute changes. GAP: We developed and validated NeoPose, a low-cost, non-invasive, computer vision digital health tool to continuously monitor neonates using exclusively video data, allowing us to accurately confirm the presence of cerebral dysfunction associated with future neurodevelopmental impairment, previously only diagnosed by electroencephalography and neuroimaging. HYPOTHESIS: We hypothesize that NeoPose, our neuromonitoring AI, can accurately identify infant anatomic landmarks (primary endpoint) and predict sedation and cerebral dysfunction (secondary) from video data alone on the largest and only multi-site video datasets and the first prospective deployment of video AI in pediatric care. METHODS: We will (1) use 4,022 videos (86% inpatient) from 2,822 infants to optimize NeoPose for edge device computation, (2) use 2.2x10-6 video minutes of 500 infants from three NICUs with video-EEG data to retrospectively validate NeoPose for identifying cerebral dysfunction and sedation, and (3) prospectively evaluate NeoPose on bedside iPhone video streams in 50 infants. RESULTS: Pending. IMPACT: NeoPose fills significant gaps: the scarcity of FDA-approved neonatal devices, lack of AI in pediatrics, and absence of effective tools for diagnosing/managing debilitating neurological conditions. With NeoPose, our validated computer vision algorithm coupled with edge device video streams, we will empower bedside clinicians with real-time AI insights to improve outcomes for the most vulnerable patients worldwide.