PosterPDF Available

Improvement of a convulsive seizure detector relying on accelerometer and electrodermal activity collected continuously by a wristband



Wrist acceleration (ACM), measured through a wearable device, has been used to automatically identify the motion of convulsive seizures such as generalized tonic-clonic seizures (GTCSs). ACM-based seizure detectors can trigger with many repetitive movements, and are affected by high false alarm rates, which might prevent daily life use. Recently, EDA and ACM features have been used to improve the classification performance of a seizure detector, reducing the false alarm rate in the detection of 16 GTCSs recorded from 7 patients. Here we present the results of further improvements, better matched to the real-time computational capacity of wearable devices, building a seizure detector from a higher number of patients and GTCSs (i.e., 38 GTCSs from 18 patients).
Acceleration (ACM), measured through a wrist-band, has been used to
automatically identify the motion of convulsive seizures such as generalized
tonic-clonic seizures (GTCSs) [1] [2]. However, ACM-based seizure
detectors have high false alarm rates, which might prevent daily life use.
Electrodermal activity (EDA) is a physiological signal reflecting the activity
of the sweat glands driven by the Sympathetic Nervous System. Direct
stimulation of some subcortical regions [3] elicits ipsilateral EDA responses.
Using both EDA and ACM features improves the classification performance
of a GTCS detector, reducing the false alarm rate [4] (16 GTCSs from 7
Using a larger feature set further lowers the false alarm rate while
preserving high sensitivity [5] (20 GTCSs from 9 patients).
Here we present the results of further improvements, using a smaller set of
features, better matched to use in patient-friendly wearable devices: 38
GTCSs from 18 patients, obtaining similarly high sensitivity (92-95%) with
low false alarm rates (0.56-2.26 per day).
F Onorati1, G Regalia1, C Caborni1, R Picard1,2*
1. Empatica, Inc, Cambridge, MA and Milan, Italy, 2. MIT Media Lab, Massachusetts Institute of Technology, Cambridge, MA
*emails: or
[1] J Lockman et al., “Detection of seizure-like movements using a wrist accelerometer”,
EpilepsyandBehavior, v. 20, no. 4, pp. 638-641, 2011.
[2] S. Beniczky et al., “Detection of generalized tonic–clonic seizures by a wireless wrist accelerometer: A
prospective, multicenter study”, Epilepsia, v. 54, no. 4, pp. e58–e61, 2013.
[3] C. A. Mangina and J H Beuzeron-Mangina, “Direct electrical stimulation of specific human brain
structures and bilateral electrodermal activity”, IntJofPsychophysiology, v. 22, no. 1–2, pp. 1–8, 1996.
[4] M-Z. Poh et al., “Convulsive seizure detection using a wrist-worn electrodermal activity and
accelerometry biosensor,” Epilepsia, v. 53, no. 5, pp. e93–e97, 2012.
[5] G. Regalia et al., “An improved wrist-worn convulsive seizure detector based on accelerometry and
electrodermal activity sensors”, AmericanEpilepsySocietyannualmeeting 2015, abs no. 3096, 2015.
Labeled seizure data were collected during clinical video
EEG (v-EEG) monitoring.
Data: 44 recordings from 18 patients wearing a wrist-band
able to record EDA and 3-axes ACM (Figure 1(a)).
EDA and ACM signals were analyzed off-line using
proprietary software to clean the data and extract signal
features over a 10 sec. window every 2.5 sec (overlap: 75%).
Support Vector Machine (SVM) classifiers were trained: one
with 46 features (SVM_46), one with 30 features (SVM_30).
A leave-one-patient-out cross-validation approach was used
to test classifier sensitivity (Se) and false alarm rate (FAR),
defined as number of false alarms per day.
The optimal decision threshold was selected by receiver
operating characteristic (ROC) curve analysis.
Recordings included 38 GTCSs from 18 patients over a total
of 1027 hours (42.7 days). Both classifiers show acceptable
FAR while keeping Se higher than 90% (Figure 2).
SVM_46 at Se=92% (i.e., 35 seizures detected out of 38)
showed FAR=0.56 and at Se=95% (i.e., 36 out of 38) FAR=2.26.
SVM_30 achieved similar performance (Se=92% and FAR=0.74,
Se=95% and FAR=2.02), using fewer features.
A seizure detection system based on ACM and EDA features was
developed using clinical data collected from a larger number of patients and
seizures with respect to previous work, capturing greater subject variability of
GTCS expression. The classifier we obtained allows a higher seizure
detection rate while maintaining an acceptable false alarm rate.
Furthermore, it is efficiently integrated into a wearable wristband to provide
real-time alarms of ongoing seizures.
teams at Emory and Childrens Healthcare of Atlanta Hospital,Atlanta, for their generous support collecting
Figure 1. (a) EDA and 3-axes ACM signals of two patients recorded
during a generalized tonic-clonic seizure (GTCS) with a wrist-worn
device. (b) Schematic workflow of the GTCSs detector.
Figure 2. Performance of the SVM classifier based on 46 EDA and ACM
features (SVM_46, blue line) and re-trained by using 30 features (SVM_30,
red line).
Figure 3. Distribution of FAR (top) and Sensitivity (bottom) of the new
SVM_30 GTCS detector across data from 18 patients (38 GTCSs, 1027
hours of recordings).
When using SVM_30 with a threshold set to provide Se 95%,
most of the patients (10/18 = 55%) had less than 1 false
alarm every 2 days and most (16/18 = 88%) had ALL of their
GTCSs detected (Figure 3).
Improvement of a convulsive seizure detector relying on
accelerometer and electrodermal activity collected continuously by a wristband
... Seizure and fall detection are generally performed separately. The gold standard for seizure detection in a clinical setting utilizes video-electroencephalography (video-EEG) monitoring [22,23]. Whilst accurate, this is costly and impractical for use in everyday life, due to its greatly limited range and physical constraints, requiring test subjects to wear intrusive sensors on their scalp. ...
... There are reports of highly accurate seizure detection systems using dedicated wristworn devices that have been helped by using electrodermal activity (EDA) sensors [22,25,26], which are currently unavailable in mainstream wearable devices. ...
... For instance, older smartphone devices might not feature accelerometers with optimal characteristics for detecting falls. Moreover, accurate heartrate and EDA sensors are not commonly found in smartwatch devices, even though their inclusion would be beneficial for detecting alarming states [22,25]. Nevertheless, with the increasing quantity and quality of sensors in modern devices, this is becoming less of an issue. ...
Conference Paper
Epilepsy and falls incur a great social and economic cost globally. Automatically detecting their occurrence would help mitigate the myriad of issues that arise from not receiving assistance after such an event. Despite existing research showing the potential advantages in using the ever-improving sensor technology incorporated within commercially available smartphone and smartwatch devices for human activity recognition, most available solutions for fall and seizure detection are still offered with dedicated hardware, which is often more expensive and less practical. This paper presents a comparison and evaluation of algorithms for detecting convulsions and falls, separately and combined, using smartphone and smartwatch devices. With a dataset of ordinary activities and simulated falls and convulsions, recorded by 15 test subjects, we found the devices a viable option for the successful detection of the activities, achieving accuracy rates between 89.7% and 98.5% with C4.5 decision tree algorithms.
... Beyond assessment of behavioral and physiological metrics, sensor technology might also be more readily involved in the prediction of onset or relapse in EDs. Multiple studies have used wearable sensor technology to predict onset of health behaviors, such as opioid usage, relapse from depression, and epileptic seizures (Onorati, Regalia, Caborni, & Picard, 2016;Saeb et al., 2015). For example, wearable sensor technology, such as Empatica wristbands, are a noninvasive method to assess EDA, acceleration, and blood-volume pulse in real time. ...
... For example, wearable sensor technology, such as Empatica wristbands, are a noninvasive method to assess EDA, acceleration, and blood-volume pulse in real time. Empatica sensor technology can be used to predict the onset of seizures in epilepsy with 77% accuracy using acceleration and EDA data (Onorati et al., 2016). Research supports a correlation among acceleration, EDA, and ED behaviors (e.g., Alberti et al., 2013), suggesting that it is possible to use wearable technology to predict the onset of ED behaviors (e.g., binge eating), which could then be used to trigger in-the-moment alerts to interrupt such behaviors, ultimately detecting and preventing ED behaviors and ultimately relapse. ...
Sensor technology has made huge technological advances in the past decade. Many sensor technologies (e.g., wearable wristbands) have been integrated into health research with the ability to substantially improve health outcomes and reduce health care costs. Despite the rapid technological developments in sensor technology, little research has examined sensor technology in eating disorders (EDs). The overarching aim of the current article is to briefly review the literature on sensor technology and health outcomes, including EDs, and discuss several potential ideas for the application of sensor technology in the treatment, assessment, and diagnosis of EDs. We will also present data from a feasibility case study with an ED participant and healthy control providing a brief example of how wearable sensor technology might be implemented in ED research. Overall, we will discuss how sensor technology could be used to improve treatment and assessment of EDs and represents an idea in need of more research in the ED field.
... Lack of GPS location and polysomnographic or concurrent EEG recording makes analysis of nocturnal seizures challenging. Future work would benefit from the use of wearable biosensors that increases certainty about presence and absence of events at any time of day or night, as well as information about sleep time [13]. In addition to other risk factors, confounders were not accounted for in the survival analysis, so that differences in survival between risk groups may be partially attributable to baseline differences in demographic or seizure characteristics. ...
Purpose This study aims to characterize the natural history of generalized motor seizures through longitudinal stratification of patient-reported clinical seizures into high, medium and low rates of generalized motor seizures (also known as generalized tonic–clonic seizures or GTCs). Methods From 2007 to 2018, 1.4 million seizures were recorded by 12,402 users that met inclusion/exclusion criteria. The number of GTCs per year since the first seizure diary entry was calculated for each user and categorized as: Low (0 GTCs/year), Medium (1–2 GTCs/year), or High (>3 GTCs/year) GTC rates. Results Kaplan–Meier survival curves for the time until exiting the initial category were computed. There was a global difference between risk groups (p < 0.01). Further pairwise log rank tests revealed a difference between each pair of risk groups (p < 0.01). At 3 years, 40.8% of people initially presenting with high GTC rates remained in their initial category, while 77.3% of people initially presenting with low GTC rates remained in their initial category. Conclusion A patient with a low rate of GTCs is likely to remain at low risk for future GTCs, whereas higher GTC rate patients (at least one GTC/year) may leave their initial risk stratification. Thus, yearly re-assessment may be prudent when considering risk of further GTCs. Given the association between higher yearly rates of GTCs with increased SUDEP risk and morbidity in epilepsy, further validation of these findings is important for prognostication.
... Recent research shows an increased interest in unobtrusive sensing of interactive experiences [29,42]. While research on technology driven experiences and gaming is very well established, physiological analysis and unobtrusive sensing in-the-wild is a growing area of interest in affective computing and interaction design [11,35,36]. However, there is still a scarcity of emotional understanding of the usage of interactive applications used by younger audiences in-situ. ...
Conference Paper
Full-text available
While gaming and storytelling are considered to be common approaches to engage audiences with a museum’s collections, a formal comparison of the two has not been found in literature. While gaming and storytelling are considered to be common approaches to engage audiences with a museum’s collections, a formal comparison of the two has not been found in literature. In this paper, we present the design and comparative study of two distinct interventions, namely a mobile game and a mobile story that were designed to engage a young audience with the exhibit of the local natural history museum. Focusing on the same scientific content derived from the museum’s collection, we compare the effects of both interactive experiences on a group of children. When comparing engagement, enjoyment and learning outcomes, we correlate results with data derived from observations and skin conductance biofeedback. The data collected so far suggest that children are 27% more excited when using the game application compared with the story driven one. Moreover, we find that children’s excitement peaks when encountering selected artefacts presented in the museum exhibit. Finally, children’s learning nearly doubled (44%) when using the game based experience versus the story. We conclude the paper by discussing the implications of our findings and by proposing potential future improvements.
Full-text available
Precision Medicine implies a deep understanding of inter-individual differences in health and disease that are due to genetic and environmental factors. To acquire such understanding there is a need for the implementation of different types of technologies based on artificial intelligence (AI) that enable the identification of biomedically relevant patterns, facilitating progress towards individually tailored preventative and therapeutic interventions. Despite the significant scientific advances achieved so far, most of the currently used biomedical AI technologies do not account for bias detection. Furthermore, the design of the majority of algorithms ignore the sex and gender dimension and its contribution to health and disease differences among individuals. Failure in accounting for these differences will generate sub-optimal results and produce mistakes as well as discriminatory outcomes. In this review we examine the current sex and gender gaps in a subset of biomedical technologies used in relation to Precision Medicine. In addition, we provide recommendations to optimize their utilization to improve the global health and disease landscape and decrease inequalities.
Wearable automated seizure detection devices offer a high potential to improve seizure management, through continuous ambulatory monitoring, accurate seizure counts, and real-time alerts for prompt intervention. More importantly, these devices can be a life-saving help for people with a higher risk of sudden unexpected death in epilepsy (SUDEP), especially in case of generalized tonic-clonic seizures (GTCS). The Embrace and E4 wristbands (Empatica) are the first commercially available multimodal wristbands that were designed to sense the physiological hallmarks of ongoing GTCS: while Embrace only embeds a machine learning-based detection algorithm, both E4 and Embrace devices are equipped with motion (accelerometers, ACC) and electrodermal activity (EDA) sensors and both the devices received medical clearance (E4 from EU CE, Embrace from EU CE and US FDA). The aim of this contribution is to provide updated evidence of the effectiveness of GTCS detection and monitoring relying on the combination of ACM and EDA sensors. A machine learning algorithm able to recognize ACC and EDA signatures of GTCS-like events has been developed on E4 data, labeled using gold-standard video-EEG examined by epileptologists in clinical centers, and has undergone continuous improvement. While keeping an elevated sensitivity to GTCS (92-100%), algorithm improvements and growing data availability led to lower false alarm rate (FAR) from the initial ˜2 down to 0.2-1 false alarms per day, as showed by retrospective and prospective analyses in inpatient settings. Algorithm adjustment to better discriminate real-life physical activities from GTCS, has brought the initial FAR of ˜6 on outpatient real life settings, down to values comparable to best-case clinical settings (FAR < 0.5), with comparable sensitivity. Moreover, using multimodal sensing, it has been possible not only to detect GTCS but also to quantify seizure-induced autonomic dysfunction, based on automatic features of abnormal motion and EDA. The latter biosignal correlates with the duration of post-ictal generalized EEG suppression, a biomarker observed in 100% of monitored SUDEP cases.
Full-text available
In this work, the performance of an automated seizure detection system based on ACM and EDA features measured from the wrist was presented using clinical data collected from a total of 53 patients having two types of seizures. The classifier we tested allows a high seizure detection rate for GTC and FOCM seizures that had never been trained by the model (Se=100% on 12 new seizures) while maintaining an acceptable false alarm rate of 0.93/day on average over 164 days. Furthermore, it is efficiently integrated into a hardware platform to provide real-time alarms of seizures while they are occurring. In the future, the model will be tested on data collected outside the clinic, where the test conditions are expected to be much more challenging.
ResearchGate has not been able to resolve any references for this publication.