A team of clinicians, scientists, and engineers at Mount Sinai trained a deep learning posture recognition algorithm on video feeds of infants in the neonatal intensive care unit (NICU) to accurately track babies’ movements and identify key neurological indicators.
Here are the results of this new artificial intelligence (AI)-based tool, announced on November 11: Lancet’s eClinical Medicine; It could lead to a minimally invasive, scalable method for continuous neurological monitoring in the NICU and provide important real-time insights into infant health that were not previously possible.
Each year, more than 300,000 newborns are admitted to NICUs across the United States. Infant alertness is considered the most sensitive part of the neurological examination, reflecting the integrity of the entire central nervous system. Neurological deterioration in the NICU can occur unexpectedly and have devastating consequences. However, unlike cardiopulmonary telemetry, which continuously monitors a baby’s heart and lung function in the NICU, neurotelemetry remains elusive in most NICUs, despite decades of work on electroencephalography (EEG) and specialized neurological NICUs. It remains. Neurological status is assessed intermittently using physical examination, which can be inaccurate and miss subacute changes.
The Mount Sinai team hypothesized that a computer vision method that tracks infant movements could predict neurological changes in the NICU. “Pose AI” is a machine learning method that tracks anatomical landmarks in video data. It revolutionized locomotion and robotics.
The Mount Sinai team trained the AI algorithm on more than 16,938,000 seconds of video footage taken from a diverse group of 115 infants in Mount Sinai Hospital’s NICU receiving continuous video EEG monitoring. They demonstrated that Pose AI can accurately track infant landmarks in video data. They then used anatomical landmarks from the video data to predict two important conditions, sedation and brain dysfunction, with high accuracy.
“Many neonatal intensive care units have video cameras, but to date they have not applied deep learning to monitor patients,” said Felix Richter, MD, senior author of the paper and instructor in pediatric neonatal medicine at Mount Sinai. . “Our study shows that applying AI algorithms to cameras that continuously monitor infants in the NICU is an effective way to detect neurological changes earlier, potentially enabling faster intervention and better outcomes. “
The research team was surprised by how well Pose AI worked under different lighting conditions (day, night, baby receiving light therapy) and from different angles. They were also surprised to find that the Pose AI movement index was associated with both gestational age and postnatal age.
“It is important to note that this approach does not replace critical physician and nursing assessments in the NICU. Rather, it augments these assessments by providing ongoing readings that can be acted upon in a given clinical situation,” explains Dr. Richter. I did it. . “We envision a future system in which cameras continuously monitor infants in the NICU through AI, providing neural telemetry strips similar to heart rate or respiration monitoring, with alerts of changes in sedation level or brain dysfunction, along with alerts to clinicians. “Provides an intuitive and easily interpretable tool for bedside management by being able to review video and AI generated insights when needed.”
The research team noted limitations of the study, including that the AI model was trained on data collected from a single institution. This means that this algorithm and its neurological predictions must be evaluated against video data from other organs and video cameras. The research team plans to develop clinical trials to test this technology in additional NICUs and evaluate its impact on care. They are also exploring applications for other neurological disorders and expanding its use to the adult population.
“At Mount Sinai, we are committed to investigating and leveraging new artificial intelligence possibilities to ensure advanced care for our patients,” said Girish N. Nadkarni, MD, MPH, director of Mount Sinai and director of data-driven and digital medicine systems. . Director of the Clinical Intelligence Center, Charles Bronfman Institute for Personalized Medicine, and co-author of the study. “AI tools are already advancing clinical care across the Mount Sinai health system, including reducing length of stay, reducing hospital readmissions, helping target cancer diagnosis and treatment, and providing real-time care to patients based on physiological data generated by wearables. “To name just a few, we are excited to now bring this non-invasive, safe, and effective AI tool to the NICU to improve outcomes for our smallest and most vulnerable patients.”
