Abstracts

Rationale:
Neonatal seizures are a common emergency in the neonatal intensive care unit (NICU). The identification of seizures in neonates is not trivial, however, and clinical recognition of seizures is based on observation of stereotypical movements, which can be subtle and difficult to identify. Thus, long-term video EEG (vEEG) monitoring is the gold standard for determining if a neonate is having seizures. Seizure identification is a complex labor- and time-intensive task. When reviewing long-term EEG for seizures, it takes skilled technicians to review and edit the EEG record for physician review. Unfortunately, there is not always the staff with the requisite expertise to review the vEEG. One possible solution to the lack of experienced epileptologists is to train computers to classify the EEG. Recent advances in machine learning (ML) methods and computers are making it feasible to classify complex data sets, which may make it possible to identify seizures with minimal human interaction. The present study employs a series of convolutional neural networks to classify the EEG data from infants and neonates as either seizure or not seizure.
Method:
EEG data was clipped from the long-term monitoring of 10 pediatric patients who were evaluated at the NICU and/or EMU at Dell Children’s Medical Center in Austin, TX. Patients ranged from 3 days to 6 months of age. Seizures were identified and labeled by experienced pediatric epileptologists. The EEG was collected using the International 10-20 system and sampled at 256 Hz. The EEG was digitally band-pass filtered from 1 to 40 Hz offline and resampled at 128 Hz. Seizure clips were then re-clipped to four second clips for classification. The AI model was built in Python 3.6 using the Keras API and the Tensorflow-GPU platform. The AI architecture was a series of convolutional neural networks followed by densely connected layers and sigmoid output.
Results:
High model accuracy was attained for training and validation datasets (see Figure 1). The validation accuracy for the model is 95%. In addition, the model loss function converged toward zero (see Figure 2).
Conclusion:
The results of this feasibility study suggest that it may be possible to train computers to classify neonatal and infant EEGs to identify seizures in a NICU or EMU setting. The use of AI to identify seizures could aid the NICU staff in identifying seizures in their patients and improve treatment response times for seizures. This study demonstrates that it is possible to use AI to classify seizures in infants and neonates. Future research needs to use a larger data set to determine if the AI model can be generalized to the population at large.
Funding:
:No funding was received for this study