PFI-TT: Artificial Intelligence-enabled Real-time System for Early Epileptic Seizure Detection and Prediction

The broader impact/commercial potential of this Partnerships for Innovation - Technology Translation (PFI-TT) project will be in the area of seizure prediction for patients suffering from drug-resistant epilepsy. The technology developed can have impact on more than 3.4 million Americans that suffer from epilepsy – including 1 million who suffer from drug-resistant epilepsy.

Current infrastructure available to the epileptic population is inadequate and is mostly reactive i.e., support is provided after a seizure attack. Physical injury, social ostracization (emotional injury), or limited opportunities (economic injury) result from the inadequate ability to predict siezures. The proposed Artificial Intelligence (AI)-based models will be incorporated into wearable sensors that detect abnormalities in brain electrical activity.

The technology will be incorporated into devices like smart phones, and will be non-invasive, and low-cost. Siezure prediction can enable timely human mitigation measures thus providing value by reducing emergency room costs, improving quality of life, and allowing caregivers to provide precautionary measures such as anti-seizure medications. With recent advances in seizure rescue therapeutics, the proposed early prediction technology can help patients make better decisions on when to medicate to prevent a seizure.

The proposed project will design and develop advanced machine learning algorithms to identify neuromarkers that can be used for the prediction of epileptic seizures using data from wearable electroencephalography (EEG). The goal of this project is to provide computational infrastructure that can predict seizures with high sensitivity and low false positive rates, and can provide real-time continuous monitoring making it highly impactful for patients and caregivers.

These solutions will be developed by formulating deep-learning models that will combine residual and long-short term memory(LSTM) layers for feature extraction for improved sensitivity and specificity for class imbalances. This development will be followed by prediction using fully connected layers. To ensure generalizability, the models will be trained and tested using data from various EEG data acquisition sites and techniques.

The edge/federated computing infrastructure will be formulated to alert patients and caregivers to take preventative measures about an impending seizure resulting in better outcomes for the patients.

This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.