VJNeurology is committed to improving our service to you

Share this video  

VJNeurology is committed to improving our service to you

EEC 2022 | Trialing ultra-long-term subcutaneous EEG for seizure forecasting in epilepsy

Pedro Viana, MD, PhD Candidate, King’s College London, London, UK, shares the results of two cohorts of patients with epilepsy that trialed minimally invasive ultra-long-term subcutaneous EEG (ULT-EEG) for seizure forecasting. The ULT-EEG consists of an implant under the skin on one side of the brain and an external data logger which connects over the skin via induction, to record data over many weeks. Subcutaneous ULT-EEG is also advantageous compared to more invasive EEG systems, such as intercranial EEG where electrodes are placed inside the brain and holds more risk for patients. Dr Viana explains that his lab trained deep learning models on the obtained data, specifically the preictal and interictal EEG segments, to tailor models for each patient and test these models at different time points during the day. They found promising results, as forecasts were statistically significant in 3-5/6 patients, depending on the model used. The forecasting system was able to predict 64-80% of seizures, while keeping the time in warning (i.e., where the forecast is in a high warning state) as low as possible, around 10-44% of the whole day. This interview took place at the European Epilepsy Congress (EEC) 2022.

Transcript (edited for clarity)

We’re going to present some results from two center cohorts. So, these were two studies separately. One is still ongoing in our center. The other center is in Denmark from the Zealand University Hospital. And in both of these studies, patients with epilepsy with frequently enough seizures, so around at least two seizures reported per month, were asked to monitor, from home, their own epilepsy...

We’re going to present some results from two center cohorts. So, these were two studies separately. One is still ongoing in our center. The other center is in Denmark from the Zealand University Hospital. And in both of these studies, patients with epilepsy with frequently enough seizures, so around at least two seizures reported per month, were asked to monitor, from home, their own epilepsy. So, to self report their seizures, either on paper or on an electronic diary, while at the same time using this new device, which is ultra long-term subcutaneous EEG. And just to briefly explain this device, this is a minimally invasive device. So, it consists of two parts. One is an implant, which is a 10 centimeter wire with a coin shaped housing that houses the electronics. And all of this is placed under the skin. And with this device, this is under local anesthesia and it’s placed on one side, so unilaterally. And we place this implant according to the region where we believe the seizures are best seen on the EEG. And then patients use an external device, a data logger, that connects over the skin via induction, and it starts recording.

So, people can record EEG for many weeks in quite an unobtrusive manner. And this is clearly an advantage over more invasive EEG systems, such as intercranial EEG where the electronics are actually placed inside the brain or at the surface of the brain and this of course carries more risks to patients. So, we believe that this might be a solution that applies to a wider population of patient that would be more acceptable. And so we used data from these two cohorts and then we annotated every seizure that we could find on the EEG for each of these patients. And we trained some deep learning models on these data and we actually trained on the hour preceding the onset of the seizure so that we could devise some kind of seizure forecasting system that could be used in real life. And then we tested this on a separate part of the recording for each of these patients.

So, each patient had its own set of models, so they were tailored to the specific patient. We found quite promising results. So, these are early results were quite promising. We used three different types of models on these data for each patient. And we found that the forecasts were statistically significant, so they were better than a random predictor in three to five out of the six patients, depending on each model. And the performance results were, I would say, modest at the beginning, but they’re still quite promising. So, we were able to, for example, predict from 64% to 80% of the seizures in these significant forecasts, while at the same time holding the time in warning, so, this will be the time where the forecast will be at a high warning stage, as low as possible. And this was at the range of 10%, 10% of the whole day, up to 44%, depending on the patient and on each model. So, again, quite promising results still early phase. We are not sure how this will translate clinically in the future, but this is clearly a good start.

Read more...

Disclosures

P.F.V. received a payment from UNEEG medical for data annotation in an unrelated research study.