Nasseri, Mona

Objective: To evaluate the feasibility and effectiveness of wrist-worn wearable devices combined with machine learning (ML) approaches for detecting a diverse array of seizure types beyond generalized tonic–clonic (GTC), including focal, generalized, and subclinical seizures. Materials and Methods: Twenty-eight patients undergoing inpatient video-EEG monitoring at Mayo Clinic were concurrently monitored using Empatica E4 wrist-worn devices. These devices captured accelerometry, blood volume pulse, electrodermal activity, skin temperature, and heart rate. Seizures were annotated by neurologists. The data were preprocessed to experiment with various segment lengths (10 s and 60 s) and multiple feature sets. Three ML strategies, XGBoost, deep learning models (LSTM, CNN, Transformer), and ROCKET, were evaluated using leave-one-patient-out cross-validation. Performance was assessed using area under the receiver operating characteristic curve (AUROC), seizure-wise recall (SW-Recall), and false alarms per hour (FA/h). Results: Detection performance varied by seizure type and model. GTC seizures were detected most reliably (AUROC = 0.86, SW-Recall = 0.81, FA/h = 3.03). Hyperkinetic and tonic seizures showed high SW-Recall but also high FA/h. Subclinical and aware-dyscognitive seizures exhibited the lowest SW-Recall and highest FA/h. MultiROCKET and XGBoost performed best overall, though no single model was optimal for all seizure types. Longer segments (60 s) generally reduced FA/h. Feature set effectiveness varied, with multi-biosignal sets improving performance across seizure types. Conclusions: Wrist-worn wearables combined with ML can extend seizure detection beyond GTC seizures, though performance remains limited for non-motor types. Optimizing model selection, feature sets, and segment lengths, and minimizing false alarms, are key to clinical utility and real-world adoption.

AbstractObjectiveEpilepsy management employs self‐reported seizure diaries, despite evidence of seizure underreporting. Wearable and implantable seizure detection devices are now becoming more widely available. There are no clear guidelines about what levels of accuracy are sufficient. This study aimed to simulate clinical use cases and identify the necessary level of accuracy for each.MethodsUsing a realistic seizure simulator (CHOCOLATES), a ground truth was produced, which was then sampled to generate signals from simulated seizure detectors of various capabilities. Five use cases were evaluated: (1) randomized clinical trials (RCTs), (2) medication adjustment in clinic, (3) injury prevention, (4) sudden unexpected death in epilepsy (SUDEP) prevention, and (5) treatment of seizure clusters. We considered sensitivity (0%–100%), false alarm rate (FAR; 0–2/day), and device type (external wearable vs. implant) in each scenario.ResultsThe RCT case was efficient for a wide range of wearable parameters, though implantable devices were preferred. Lower accuracy wearables resulted in subtle changes in the distribution of patients enrolled in RCTs, and therefore higher sensitivity and lower FAR values were preferred. In the clinic case, a wide range of sensitivity, FAR, and device type yielded similar results. For injury prevention, SUDEP prevention, and seizure cluster treatment, each scenario required high sensitivity and yet was minimally influenced by FAR.SignificanceThe choice of use case is paramount in determining acceptable accuracy levels for a wearable seizure detection device. We offer simulation results for determining and verifying utility for specific use case and specific wearable parameters.

Abstract Objective The factors that influence seizure timing are poorly understood, and seizure unpredictability remains a major cause of disability. Work in chronobiology has shown that cyclical physiological phenomena are ubiquitous, with daily and multiday cycles evident in immune, endocrine, metabolic, neurological, and cardiovascular function. Additionally, work with chronic brain recordings has identified that seizure risk is linked to daily and multiday cycles in brain activity. Here, we provide the first characterization of the relationships between the cyclical modulation of a diverse set of physiological signals, brain activity, and seizure timing. Methods In this cohort study, 14 subjects underwent chronic ambulatory monitoring with a multimodal wrist‐worn sensor (recording heart rate, accelerometry, electrodermal activity, and temperature) and an implanted responsive neurostimulation system (recording interictal epileptiform abnormalities and electrographic seizures). Wavelet and filter–Hilbert spectral analyses characterized circadian and multiday cycles in brain and wearable recordings. Circular statistics assessed electrographic seizure timing and cycles in physiology. Results Ten subjects met inclusion criteria. The mean recording duration was 232 days. Seven subjects had reliable electroencephalographic seizure detections (mean = 76 seizures). Multiday cycles were present in all wearable device signals across all subjects. Seizure timing was phase locked to multiday cycles in five (temperature), four (heart rate, phasic electrodermal activity), and three (accelerometry, heart rate variability, tonic electrodermal activity) subjects. Notably, after regression of behavioral covariates from heart rate, six of seven subjects had seizure phase locking to the residual heart rate signal. Significance Seizure timing is associated with daily and multiday cycles in multiple physiological processes. Chronic multimodal wearable device recordings can situate rare paroxysmal events, like seizures, within a broader chronobiology context of the individual. Wearable devices may advance the understanding of factors that influence seizure risk and enable personalized time‐varying approaches to epilepsy care.

Abstract This study describes a generalized cross‐patient seizure‐forecasting approach using recurrent neural networks with ultra‐long‐term subcutaneous EEG (sqEEG) recordings. Data from six patients diagnosed with refractory epilepsy and monitored with an sqEEG device were used to develop a generalized algorithm for seizure forecasting using long short‐term memory (LSTM) deep‐learning classifiers. Electrographic seizures were identified by a board‐certified epileptologist. One‐minute data segments were labeled as preictal or interictal based on their relationship to confirmed seizures. Data were separated into training and testing data sets, and to compensate for the unbalanced data ratio in training, noise‐added copies of preictal data segments were generated to expand the training data set. The mean and standard deviation (SD) of the training data were used to normalize all data, preserving the pseudo‐prospective nature of the analysis. Different architecture classifiers were trained and tested using a leave‐one‐patient‐out cross‐validation method, and the area under the receiver‐operating characteristic (ROC) curve (AUC) was used to evaluate the performance classifiers. The importance of each input signal was evaluated using a leave‐one‐signal‐out method with repeated training and testing for each classifier. Cross‐patient classifiers achieved performance significantly better than chance in four of the six patients and an overall mean AUC of 0.602 ± 0.126 (mean ± SD). A time in warning of 37.386% ± 5.006% (mean ± std) and sensitivity of 0.691 ± 0.068 (mean ± std) were observed for patients with better than chance results. Analysis of input channels showed a significant contribution ( p < .05) by the Fourier transform of signals channels to overall classifier performance. The relative contribution of input signals varied among patients and architectures, suggesting that the inclusion of all signals contributes to robustness in a cross‐patient classifier. These early results show that it is possible to forecast seizures training with data from different patients using two‐channel ultra‐long‐term sqEEG.