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

Abstract

The ability to forecast seizures minutes to hours in advance of an event has been verified using invasive EEG devices, but has not been previously demonstrated using noninvasive wearable devices over long durations in an ambulatory setting. In this study we developed a seizure forecasting system with a long short-term memory (LSTM) recurrent neural network (RNN) algorithm, using a noninvasive wrist-worn research-grade physiological sensor device, and tested the system in patients with epilepsy in the field, with concurrent invasive EEG confirmation of seizures via an implanted recording device. The system achieved forecasting performance significantly better than a random predictor for 5 of 6 patients studied, with mean AUC-ROC of 0.80 (range 0.72–0.92). These results provide the first clear evidence that direct seizure forecasts are possible using wearable devices in the ambulatory setting for many patients with epilepsy.

Multimodal wearable sensors inform cycles of seizure risk

“…The assessment of seizure risk in epilepsy may be improved further by combining long timescale cycles based seizure risk characterization with an acute seizure forecast layer, which has shown early promise with multimodal wearable devices. 46 More work is needed to optimize the performance of multimodal wearable sensors, however, the results from our straightforward approach are encouraging. Even if multimodal wrist-worn devices are unable to attain the performance of a brain implant, the ease of implementation, lower cost, broad applicability, and patient preference for a wrist-worn form factor 49 make this a critical area of study.…”

“…The device’s on-board memory is cleared after each upload. As previously described 46 electrographic seizures were verified by a board certified epileptologist (N.M.G) and experienced reader (J.B.). Seizure analyses were limited to subjects for whom ‘long-episodes’ available for visual confirmation did not exceed the device’s storage capacity except rarely.…”

“…1), as previously described. 4,46 The hourly rate of IEA is recorded by the implanted device, and these data were used to assess cycles in IEA activity. The RNS device stores between 60 and 180 seconds of timeseries data (a 'long-episode' in NeuroPace terminology) when the seizure detector is triggered and can store up to approximately 10 minutes of timeseries data on the implanted device.…”

Seizure Detection and Prediction by Parallel Memristive Convolutional Neural Networks

“…Seizure is a chronic, recurring condition that can mostly be prevented through medication before onset [51], but even with the best medications, 30% of the patients are drug-resistant [52]. Closed-loop brain stimulation has been found to mitigate and even improve symptoms [53], [54], but unpredictability of seizure requires a closed-loop prediction system to provide accurate warning with adequate preparation time for stimulation [55]. This calls for the need for fast, low-latency computations, as the changes within the patients can be noticed early-on, in order to start treatments early to improve safety and quality of life [56].…”

Medicom Conference Report EAN 2022

“…This can and should be accomplished by the following 7 interventions: brain and carotid imaging, neurology consultation, antihypertensive treatment, anticoagulants for atrial fibrillation, anti-thrombotic medication, and statins. A recent American study showed that only a third (34.3%) of patients actually received these interventions without fail after a TIA [3]. Epilepsy care is another example where quality-of-care measures are not sufficiently adhered to in clinical practice, as Prof. Clarke illustrated by means of some of the quality-of-care measures from the American Academy of Neurology (AAN):…”