“New AI Model Predicts Seizures in Newborns with Brain Injury”
In a groundbreaking study, researchers developed a machine learning (ML) model designed to predict the initial onset of seizures in newborns suffering from hypoxic–ischemic encephalopathy (HIE). HIE is a serious condition that affects infants due to a lack of oxygen and blood flow to the brain, and seizures are a common complication. The model, dubbed Neo-GB, leverages both clinical data (like blood gas values and gestational age) and quantitative electroencephalogram (QEEG) features to assess seizure risk over time.
To create the Neo-GB model, the researchers utilized a dataset from the University of California, San Francisco (UCSF), which included 60 neonates, half of whom experienced seizures. They trained the model using various clinical factors and specific QEEG metrics, such as statistical moments and spectral power, which reflect brain activity. They also enhanced their training with additional publicly available EEG data from other hospitals. The results were promising, showing that Neo-GB could effectively forecast seizure onset both statically at a specific time and dynamically over a 24-hour period.
The model’s performance was impressive; for static forecasts 30 minutes after EEG initiation, Neo-GB achieved an area under the receiver operating characteristic curve (AUROC) of 0.76 without time-dependent features and 0.89 with them. In dynamic assessments over 24 hours, the model showed a median iAUC of 0.79, indicating a solid ability to predict when seizures would occur. Notably, the factors that influenced the model’s predictions the most included spectral power, postmenstrual age, and cord blood gas values.
This research highlights the potential of combining advanced technology like machine learning with clinical assessments to improve outcomes for newborns with HIE. By identifying changes in spectral power in the EEG leading up to a seizure, clinicians could potentially intervene earlier and more effectively. Overall, this study opens the door for further advancements in monitoring and treating neonatal seizures, ultimately aiming to enhance the quality of care for vulnerable infants.