Home > Neurology > EAN 2022 > Epilepsy > Seizure forecasting with non- and minimally-invasive devices

Seizure forecasting with non- and minimally-invasive devices

Presented By
Dr Pedro Faro Viana, King's College London, UK
EAN 2022

An international project has established the feasibility of ambulatory seizure forecasting with multiple minimally-invasive devices, from medical-grade/research-grade devices to fitness trackers, using seizure cycles and subcutaneous EEG. Epileptic seizure cycles are common, measurable with different modalities, and strong predictors of future seizures.

From a patient’s perspective, one of the most difficult and debilitating aspects of epilepsy is seizure unpredictability. “Because of the potential harm that seizures can inflict onto patients or others, including sudden death, patients live in a state of perpetual uncertainty,” said lead author of the project Dr Pedro Faro Viana (King’s College London, UK) [1]. Forecasting of seizures with an implantable intracranial EEG (iEEG) device is not appropriate for all patients and in any case minimally- or non-invasive EEG recording systems would be preferable.

Patients with drug-resistant epilepsy were enrolled in a prospective, multicentre, observational cohort for ultra long-term (>8 months) monitoring with an electronic diary, a wearable device, and ambulatory EEG monitoring [1]. Wearable devices employed were the Empatica E4, the smaller and more stylish Embrace 2 (if the Empatica E4 was refused), and the Fitbit Charge 3 or Fitbit Inspire. Ambulatory EEG devices included were NeuroPace RNS, Medtronic RC+S, UNEEG 24/7 SubQ, and EpiMinder.

A total of 40 participants had recorded >11,400 days (31.2 years) of ambulatory data, including 1,712 EEG seizures. Nine patients terminated this observational study prematurely. These were some of the results:

  • The Empatica E4 and NeuroPace RNS showed that seizures occur in cycles. Heart rate circadian and multi-day cycles were significantly phase-locked with the likelihood of self-reported seizures [2].
  • The Empatica E4 plus a seizure forecasting system with a long short-term memory (LSTM) recurrent neural network (RNN) algorithm predicted seizures significantly better than chance in 5/6 patients, with a mean area under the curve (AUC) of 0.80 [3].
  • The wrist-worn Fitbit plus LSTM RNN also predicted seizures significantly better than chance in 5/6 patients, with an AUC of 0.74 (hourly forecasts) and 0.66 (daily forecasts) [4].
  • The UNEEG 24/7 subscalp achieved significant forecasting performance in 3–5 out of 6 patients, with overall mean AUC of 0.65–0.74 [5].

The main study limitations were a small cohort, some cases of device deficiency, poor adherence and/or incomplete data, and retrospective or (pseudo-) prospective forecasting.

  1. Viana P, et al. Seizure forecasting with non-invasive and minimally-invasive mobile devices – Epilepsy Foundation My Seizure Gauge study. OPR-078, EAN 2022, 25–28 April, Vienna, Austria.
  2. Gregg NM, et al. Under review.
  3. Nasseri M, et al. Sci Rep. 2021;11(1):21935.
  4. Stirling RE, et al. Front Neurol. 2021;12:713794.
  5. Viana PF, et al. Epilepsia. Apr 8 2022. DOI: 10.1111/epi.17252.

Copyright ©2022 Medicom Medical Publishers

Posted on