Artificial intelligence-enhanced epileptic seizure detection by wearables.

OBJECTIVE: Wrist- or ankle-worn devices are less intrusive than the widely used electroencephalographic (EEG) systems for monitoring epileptic seizures. Using custom-developed deep-learning seizure detection models, we demonstrate the detection of a broad range of seizure types by wearable signals. METHODS: Patients admitted to the epilepsy monitoring unit were enrolled and asked to wear wearable sensors on either wrists or ankles. We collected patients' electrodermal activity, accelerometry (ACC), and photoplethysmography, from which blood volume pulse (BVP) is derived. Board-certified epileptologists determined seizure onset, offset, and types using video and EEG recordings per the International League Against Epilepsy 2017 classification. We applied three neural network models-a convolutional neural network (CNN) and a CNN-long short-term memory (LSTM)-based generalized detection model and an autoencoder-based personalized detection model-to the raw time-series sensor data to detect seizures and utilized performance measures, including sensitivity, false positive rate (the number of false alarms divided by the total number of nonseizure segments), number of false alarms per day, and detection delay. We applied a 10-fold patientwise cross-validation scheme to the multisignal biosensor data and evaluated model performance on 28 seizure types. RESULTS: We analyzed 166 patients (47.6% female, median age = 10.0 years) and 900 seizures (13 254 h of sensor data) for 28 seizure types. With a CNN-LSTM-based seizure detection model, ACC, BVP, and their fusion performed better than chance; ACC and BVP data fusion reached the best detection performance of 83.9% sensitivity and 35.3% false positive rate. Nineteen of 28 seizure types could be detected by at least one data modality with area under receiver operating characteristic curve > .8 performance. SIGNIFICANCE: Results from this in-hospital study contribute to a paradigm shift in epilepsy care that entails noninvasive seizure detection, provides time-sensitive and accurate data on additional clinical seizure types, and proposes a novel combination of an out-of-the-box monitoring algorithm with an individualized person-oriented seizure detection approach.

Seizure detection using wearable sensors and machine learning: Setting a benchmark.

OBJECTIVE: Tracking seizures is crucial for epilepsy monitoring and treatment evaluation. Current epilepsy care relies on caretaker seizure diaries, but clinical seizure monitoring may miss seizures. Wearable devices may be better tolerated and more suitable for long-term ambulatory monitoring. This study evaluates the seizure detection performance of custom-developed machine learning (ML) algorithms across a broad spectrum of epileptic seizures utilizing wrist- and ankle-worn multisignal biosensors. METHODS: We enrolled patients admitted to the epilepsy monitoring unit and asked them to wear a wearable sensor on either their wrists or ankles. The sensor recorded body temperature, electrodermal activity, accelerometry (ACC), and photoplethysmography, which provides blood volume pulse (BVP). We used electroencephalographic seizure onset and offset as determined by a board-certified epileptologist as a standard comparison. We trained and validated ML for two different algorithms: Algorithm 1, ML methods for developing seizure type-specific detection models for nine individual seizure types; and Algorithm 2, ML methods for building general seizure type-agnostic detection, lumping together all seizure types. RESULTS: We included 94 patients (57.4% female, median age = 9.9 years) and 548 epileptic seizures (11 066 h of sensor data) for a total of 930 seizures and nine seizure types. Algorithm 1 detected eight of nine seizure types better than chance (area under the receiver operating characteristic curve [AUC-ROC] = .648-.976). Algorithm 2 detected all nine seizure types better than chance (AUC-ROC = .642-.995); a fusion of ACC and BVP modalities achieved the best AUC-ROC (.752) when combining all seizure types together. SIGNIFICANCE: Automatic seizure detection using ML from multimodal wearable sensor data is feasible across a broad spectrum of epileptic seizures. Preliminary results show better than chance seizure detection. The next steps include validation of our results in larger datasets, evaluation of the detection utility tool for additional clinical seizure types, and integration of additional clinical information.

Association between semiologic, autonomic, and electrographic seizure characteristics in children with generalized tonic-clonic seizures.

INTRODUCTION: Generalized tonic-clonic seizures (GTCS) are associated with elevated electrodermal activity (EDA) and postictal generalized electroencephalographic suppression (PGES), markers that may indicate sudden unexpected death in epilepsy (SUDEP) risk. This study investigated the association of GTCS semiology, EDA, and PGES in children with epilepsy. METHODS: Patients admitted to the Boston Children's Hospital long-term video-EEG monitoring unit wore a sensor that records EDA. We selected patients with at least one GTCS and reviewed video-EEGs for semiology, tonic and clonic phase duration, total clinical seizure duration, electrographic onset, offset, and PGES. We grouped patients into three semiology classes: GTCS 1: bilateral symmetric tonic arm extension, GTCS 2: no specific tonic arm extension or flexion, GTCS 3: unilateral or asymmetrical arm extension, tonic arm flexion or posturing that does not fit into GTCS 1 or 2. We analyzed the correlation between semiology, EDA, and PGES, and measured the area under the curve (AUC) of the ictal EDA (seizure onset to one hour after), subtracting baseline EDA (one-hour seizure-free before seizure onset). Using generalized estimating equation (GEE) and linear regression, we analyzed all seizures and single episodes per patient. RESULTS: We included 30 patients (median age 13.8 ±â€¯3.6 years, 46.7% females) and 53 seizures. With GEE, GTCS 1 was associated with longer PGES duration compared to GTCS 2 (Estimate (ß) = -26.32 s, 95% Confidence Interval (CI): -36.46 to -16.18, p < 0.001), and the presence of PGES was associated with greater EDA change (ß = 429604 µS, 95% CI: 3550.96 to 855657.04, p = 0.048). With single-episode analysis, GTCS 1 had greater EDA change than GTCS 2 ((ß = -601339 µS, 95% CI: -1167016.56 to -35661.44, p = 0.047). EDA increased with PGES presence (ß = 637500 µS, 95% CI: 183571.84 to 1091428.16, p = 0.01) and duration (ß = 16794 µS, 95% CI: 5729.8 to 27858.2, p = 0.006). Patients with GTCS 1 had longer PGES duration compared to GTCS 2 (ß = -30.53 s, 95% CI: -44.6 to -16.46, p < 0.001) and GTCS 3 (ß = -22.07 s, 95% CI: -38.95 to -5.19, p = 0.016). CONCLUSION: In children with epilepsy, PGES correlates with greater ictal EDA. GTCS 1 correlated with longer PGES duration and may indirectly correlate with greater ictal EDA. Our study suggests potential applications in monitoring and preventing SUDEP in these patients.

Ultradian rhythms in accelerometric and autonomic data vary based on seizure occurrence in paediatric epilepsy patients.

Ultradian rhythms are physiological oscillations that resonate with period lengths shorter than 24 hours. This study examined the expression of ultradian rhythms in patients with epilepsy, a disease defined by an enduring seizure risk that may vary cyclically. Using a wearable device, we recorded heart rate, body temperature, electrodermal activity and limb accelerometry in patients admitted to the paediatric epilepsy monitoring unit. In our case-control design, we included recordings from 29 patients with tonic-clonic seizures and 29 non-seizing controls. We spectrally decomposed each signal to identify cycle lengths of interest and compared average spectral power- and period-related markers between groups. Additionally, we related seizure occurrence to the phase of ultradian rhythm in patients with recorded seizures. We observed prominent 2- and 4-hour-long ultradian rhythms of accelerometry, as well as 4-hour-long oscillations in heart rate. Patients with seizures displayed a higher peak power in the 2-hour accelerometry rhythm (U = 287, P = 0.038) and a period-lengthened 4-hour heart rate rhythm (U = 291.5, P = 0.037). Those that seized also displayed greater mean rhythmic electrodermal activity (U = 261; P = 0.013). Most seizures occurred during the falling-to-trough quarter phase of accelerometric rhythms (13 out of 27, χ2 = 8.41, P = 0.038). Fluctuations in seizure risk or the occurrence of seizures may interrelate with ultradian rhythms of movement and autonomic function. Longitudinal assessments of ultradian patterns in larger patient samples may enable us to understand how such rhythms may improve the temporal precision of seizure forecasting models.

Development of a Multivariable Seizure Likelihood Assessment Based on Clinical Information and Short Autonomic Activity Recordings for Children With Epilepsy.

BACKGROUND: Predicting seizure likelihood for the following day would enable clinicians to extend or potentially schedule video-electroencephalography (EEG) monitoring when seizure risk is high. Combining standardized clinical data with short-term recordings of wearables to predict seizure likelihood could have high practical relevance as wearable data is easy and fast to collect. As a first step toward seizure forecasting, we classified patients based on whether they had seizures or not during the following recording. METHODS: Pediatric patients admitted to the epilepsy monitoring unit wore a wearable that recorded the heart rate (HR), heart rate variability (HRV), electrodermal activity (EDA), and peripheral body temperature. We utilized short recordings from 9:00 to 9:15 pm and compared mean values between patients with and without impending seizures. In addition, we collected clinical data: age, sex, age at first seizure, generalized slowing, focal slowing, and spikes on EEG, magnetic resonance imaging findings, and antiseizure medication reduction. We used conventional machine learning techniques with cross-validation to classify patients with and without impending seizures. RESULTS: We included 139 patients: 78 had no seizures and 61 had at least one seizure after 9 pm during the concurrent video-EEG and E4 recordings. HR (P < 0.01) and EDA (P < 0.01) were lower and HRV (P = 0.02) was higher for patients with than for patients without impending seizures. The average accuracy of group classification was 66%, and the mean area under the receiver operating characteristics was 0.72. CONCLUSIONS: Short-term wearable recordings in combination with clinical data have great potential as an easy-to-use seizure likelihood assessment tool.

Measuring Real-Time Medication Effects From Electroencephalography.

PURPOSE: Evaluating the effects of antiseizure medication (ASM) on patients with epilepsy remains a slow and challenging process. Quantifiable noninvasive markers that are measurable in real-time and provide objective and useful information could guide clinical decision-making. We examined whether the effect of ASM on patients with epilepsy can be quantitatively measured in real-time from EEGs. METHODS: This retrospective analysis was conducted on 67 patients in the long-term monitoring unit at Boston Children's Hospital. Two 30-second EEG segments were selected from each patient premedication and postmedication weaning for analysis. Nonlinear measures including entropy and recurrence quantitative analysis values were computed for each segment and compared before and after medication weaning. RESULTS: Our study found that ASM effects on the brain were measurable by nonlinear recurrence quantitative analysis on EEGs. Highly significant differences (P < 1e-11) were found in several nonlinear measures within the seizure zone in response to antiseizure medication. Moreover, the size of the medication effect correlated with a patient's seizure frequency, seizure localization, number of medications, and reported seizure frequency reduction on medication. CONCLUSIONS: Our findings show the promise of digital biomarkers to measure medication effects and epileptogenicity.

Seizure-related differences in biosignal 24-h modulation patterns.

A seizure likelihood biomarker could improve seizure monitoring and facilitate adjustment of treatments based on seizure risk. Here, we tested differences in patient-specific 24-h-modulation patterns of electrodermal activity (EDA), peripheral body temperature (TEMP), and heart rate (HR) between patients with and without seizures. We enrolled patients who underwent continuous video-EEG monitoring at Boston Children's Hospital to wear a biosensor. We divided patients into two groups: those with no seizures and those with at least one seizure during the recording period. We assessed the 24-h modulation level and amplitude of EDA, TEMP, and HR. We performed machine learning including physiological and clinical variables. Subsequently, we determined classifier performance by cross-validated machine learning. Patients with seizures (n = 49) had lower EDA levels (p = 0.031), EDA amplitudes (p = 0.045), and trended toward lower HR levels (p = 0.060) compared to patients without seizures (n = 68). Averaged cross-validated classification accuracy was 69% (AUC-ROC: 0.75). Our results show the potential to monitor and forecast risk for epileptic seizures based on changes in 24-h patterns in wearable recordings in combination with clinical variables. Such biomarkers might be applicable to inform care, such as treatment or seizure injury risk during specific periods, scheduling diagnostic tests, such as admission to the epilepsy monitoring unit, and potentially other neurological and chronic conditions.