Mount Sinai researchers have demonstrated how artificial intelligence can detect serious neurologic changes in babies using video data. The study findings serve as a foundation for broader neuro-monitoring applications across intensive care units globally.
For the study, the researchers trained a deep learning pose-recognition algorithm on video feeds of infants in the neonatal intensive care unit (NICU) to accurately track their movements and identify key neurologic metrics.
The finding could lead to a minimally invasive, scalable method for continuous neurologic monitoring in NICUs, providing new critical real-time insights into infant health.
Infant alertness is considered the most sensitive piece of the neurologic exam, reflecting integrity throughout the central nervous system. Neurologic deterioration can happen unexpectedly and unlike cardiorespiratory telemetry, which continuously monitors the heart and lung function of babies, neurotelemetry has remained elusive despite decades of work in electroencephalography (EEG).
Neurologic status is evaluated intermittently, using physical exams that are imprecise and may miss subacute changes.
The researchers hypothesized that a computer vision method to track infant movement could predict neurologic changes in the NICU. “Pose AI” is a machine learning method that tracks anatomic landmarks from video data; it has revolutionized athletics and robotics.
With this, the researchers trained an AI algorithm on more than 16,938,000 seconds of video footage from a diverse group of 115 infants in the NICU at The Mount Sinai Hospital undergoing continuous video EEG monitoring. They demonstrated that Pose AI can accurately track infant landmarks from video data. They then used anatomic landmarks from the video data to predict two critical conditions—sedation and cerebral dysfunction—with high accuracy.
The research team demonstrated that Pose AI worked effectively across different lighting conditions (day vs. night vs. in babies receiving phototherapy) and from different angles. The Pose AI movement index was associated with both gestational age and postnatal age.
There were some limitations of the study, including the AI models trained on data collected at a single institution, meaning that this algorithm and neurologic predictions need to be evaluated on video data from other institutions and video cameras.
The research appears in the journal Lancet’s eClinicalMedicine. The research is titled “Detection of neurologic changes in critically ill infants using deep learning on video data: a retrospective single center cohort study.”
