Ambulatory seizure forecasting with a wrist-worn device using long-short term memory deep learning

Area of Science:

• Biomedical Engineering
• Neuroscience
• Medical Devices

Background:

• Forecasting seizures minutes to hours in advance has been achieved with invasive electroencephalogram (EEG) devices.
• Previous attempts using noninvasive wearable devices for long-term, ambulatory seizure forecasting have not been successful.

Purpose of the Study:

• To develop and test a novel seizure forecasting system utilizing a noninvasive wearable device.
• To evaluate the system's performance in ambulatory patients with epilepsy, with concurrent invasive EEG confirmation.

Main Methods:

• Development of a seizure forecasting system employing a long short-term memory (LSTM) recurrent neural network (RNN) algorithm.
• Utilized a noninvasive, wrist-worn research-grade physiological sensor for data collection.
• Tested the system in ambulatory epilepsy patients, comparing its predictions against simultaneously recorded invasive EEG data.

Main Results:

• The LSTM-RNN based system achieved forecasting performance significantly better than random prediction for 5 out of 6 patients.
• The system demonstrated a mean Area Under the Receiver Operating Characteristic Curve (AUC-ROC) of 0.80 (range 0.72-0.92).

Conclusions:

• This study provides the first clear evidence that direct seizure forecasts are feasible using wearable devices in an ambulatory setting for many epilepsy patients.
• Noninvasive wearable technology holds promise for improving seizure management and patient quality of life.