Conference PaperPDF Available

Autonomic characterization of non-convulsive seizures based on wrist-worn sensors

Authors:

Abstract

Rationale There is a growing need for non-EEG, wearable devices to automatically monitor epileptic seizures in ambulatory settings. Efforts to date have focused on the development of devices to detect convulsive seizures, such as the FDA-approved smart watch Embrace by Empatica (Onorati et al, Epilepsia 2017, 58, 11). However, Embrace was not designed to detect non-convulsive seizures such as focal seizures with impaired awareness (FIAS). Few works have investigated the feasibility of non-convulsive seizure monitoring based on wrist-worn sensors (Thome-Souza et al 2014, AES; Poh et al, Neurology 2012, 78(23); Cogan et al, Int J Neural Syst 2017, 27, 1650031). This work characterizes autonomic changes measured at the wrist at the time of FIAS as a step toward providing a future automated monitor of non-convulsive seizures. Methods A cohort of 3 children and 9 adults (ages: 10 to 52 years, median: 37 years) experienced a total of 40 FIAS in Epilepsy Monitoring Units while wearing the Empatica E4 wristband. The E4’s sensors provided heart rate (HR) from photoplethysmography (PPG), as well as electrodermal activity (EDA) and accelerometer activity. Seizures were clinically labeled using video-EEG. A significant EDA response (EDR) after the seizure onset was detected offline, defined as the EDA level above two times the standard deviation of a 15-sec EDA baseline, drawn 1 min before the seizure. HR patterns 5 seconds before and 10 seconds after the seizure onset were analyzed in the time domain and compared to baseline values (1 min before onset). Results HR data showed an increase around the seizure onset in 80% of the seizures. The derivative of the HR in a 15-sec time window around the seizure onset was significantly higher (mean increase of 4 bpm/s, p< 0.001) than the baseline values. Overall, 11 out of 40 (28%) of the recorded FIAS showed a significant EDR (median amplitude relative increase: 181%). Interestingly, 75% of patients who usually experience both FIAS and generalized seizures exhibited an EDR with their FIAS. The peak of the EDR occurred 5 min (median value, range: 18 secs to 10 min) after the seizure onset and the EDR lasted 11 min (median value, range: 30 secs to 30 min). Conclusions We reported for the first time an analysis of HR and EDA changes related to non-convulsive seizures recorded with the E4 wristband. Most of the FIAS had a statistically significant HR increase around the seizure onset, in line with earlier observations (Cogan et al 2017). The automatic detection of significant EDR’s in our dataset was challenged by night time EDA sleep storm activation observed in the peri-ictal baseline window. The differences in cohort characteristics, sleep storm activity, and computational method may account for the fewer significantly-sized EDR’s found here than observed previously (Thome-Souza et al 2014, Poh et al 2012). A combination of HR and EDA signals could potentially be used for real-time detection of FIAS especially during rest periods. This preliminary analysis lays the groundwork for further work on a non-convulsive seizure detection and classification system with multimodal wearable devices.
Autonomic characterization of non-convulsive seizures based on wrist-worn sensors
L. Babilliot1,G.Regalia1,C.Caborni1, R.W. Picard1,2,F.Onorati1
We reported for the first time an analysis of
HR and EDA changes related to non-convulsive seizures
recorded with
the E4 wristband.
The EDA increase, despite having consistently occurred for the vast majority of the seizures, was not statistically
significant.On the other hand, the amount of EDA increase is
consistent with previous results[3]
.
HR increase is statistically
higher than the pre-ictal HR
,in line with earlier observations
[5]
.
Speculatively, the EDA increase in our dataset was limited due to:
Night time
EDA sleep storm
activations observed in the peri-ictal baseline window.
Cohort with
different demographical
and etiological characteristics with respect to the Literature
[3]
.
Very
low EDA signal
,which limited the sensitivity of the sensor and amplified the influence of movement
artifacts.
Acombination of HR and EDA signals could potentially be used for
real-time detection of FIAS
especially during rest
periods.This preliminary analysis lays the groundwork for further work on anon-convulsive seizure detection and
classification system with
multimodal wearable devices
.
Acohort of
3children
and
8adults
(ages:10 to 52 years, median:37 years) experienced atotal of
41 FIAS
in EMUs
while wearing an
Empatica E4
wristband.Seizures were clinically labeled using
video-EEG
.
The E4 provided
photoplethysmography
(PPG) and
EDA
.
HR
was obtained from PPG using a
proprietary algorithm
.
To analyze the HR, we considered 100 sec before the seizure onset as pre-ictal data and 100 sec after the seizure
offset as post-ictal data.
In the analysis of the HR trend the pre-ictal, ictal and post-ictal segments were further split in 2parts.
Regarding EDA, 60 sec before the seizure onset were considered as pre-ictal .We consider EDA to have increased if
the EDA rose higher than the pre-ictal baseline of twice the standard deviation computed in the 60 sec pre-ictal
segment (EDA>μbaseline + 2*σbaseline)in the 5minutes after the onset.
Figure 1
shows an example of EDA and HR signals before, during and after the seizure timing marked by v-EEG.
1. Empatica Inc
2. MIT Media Lab, Massachusetts Institute of Technology, Cambridge, Massachusetts, U.S.A
Figure 1-HR and EDA observed during an FIAS recorded with E4. The clinical onset and offset of the seizure are represented by the red area.
Results
To date the development of commercial non-EEG automatic detection systems have been focused mainly on
detecting convulsive seizures
Awearable smart watch monitoring seizures using
autonomic and motion data
has received
FDA approval
(Empatica Embrace)
[1][2]
.
An open challenge is to build automated systems to detect non-convulsive seizures, such as
focal with impaired awareness seizures
(FIASs).
The most promising physiological data for FIAS detection are
electrodermal activity
(EDA)
[3] [4]
and
heart rate
(HR)
[3] [4]
.
Cogan and colleagues
[5]
used EDA and HR, along with SpO2to detect FIASs.Poh et al.observed an average
EDA increase of 0.7
μ
S
during FIASs
[6]
.
Rationale Methods
Conclusions
Contacts
[1]
Onorati, F., Regalia, G., et al. (2017). Multicenter clinical assessment of improved wearable multimodal convulsive seizure detectors.
[2]
Lai et al. (2018), Performance assessment of Embrace, the first FDA approved smartwatch for convulsive seizure detection.
[3]
Van Westhrenen et al. (2018). Ictal autonomic changes as a tool for seizure detection: A systematic review.
[4]
Van de Vel et al. (2016). Non-EEG seizure-detection systems and potential SUDEP prevention: State of the art . Review and update.
[5]
Cogan et al. (2016), Multi-Biosignal Analysis for Epileptic Seizure Monitoring.
[6]
Poh et al. (2010), Continuous Monitoring of Electrodermal Activity During Epileptic Seizures Using Wearable Sensor.
References
Francesco Onorati
fo@empatica.com
Rosalind Picard
rp@empatica.com
Chiara Caborni
cc@empatica.com
Giulia Regalia
gr@empatica.com
Laurine Babilliot
lb@empatica.com
On average, the FIAS duration is
86 sec
(min:17 sec;max:532 sec).
Electrodermal Activity
Overall,
34 out of 41
(83%) of the FIASs showed an
increase in the EDA
(mean
absolute amplitude increase:
0.51
μ
S
). The observed increase is not statistically
significant (p>0.05).
The EDA response occurred on average
13 sec after the seizure onset
,and it
lasted on average 326 sec.
Figure 2
shows the EDA amplitude at different phases
of the EDA response.
Heart Rate
The
mean HR
, i.e. the average HR in the segment, increased on average by
10 bpm
after the seizure onset
.The maximum HR increase with respect to the pre-ictal
mean HR (
max
Δ
HR
)was on average
25 bpm
.
Figure 3
shows median and quartiles
of the mean HR and max ΔHR variables for pre-ictal, ictal and post-ictal segments.
One-way ANOVA analysis for both of the HR variables shows
statistically
significant differences
between groups (mean HR:F(2, 119) = 6.2183, p = 0.0027;
max ΔHR:F(2, 119) = 11.2224, p = 3.4781*10-5).
Post-hoc
analysis showed that the
differences are statistically significant
between pre-ictal and ictal
groups.
After having segmented the pre-ictal, ictal and post-ictal segments as described in
the Methods section, alinear regression of HR data in each sub-segment showed
an
HR increase in the first half of the seizure
(
Figure 4
). With an average
slope
of 0.575 bpm/sec
,the average HR increase resulted in 24.7 bpm occurring in the
first half of the seizure.
Figure 5
shows the average profile of the HR increase, delimited by 25%and 75%
percentiles.
In this contribution, we report FIAS autonomic changes as measured through a wrist-worn
device, as a step toward providing a future automated monitoring of non-convulsive seizures.
Aim
Figure 2-Median and quartiles of EDA amplitudes for
different phases of the response (N =34).
Figure 3Medians and quartiles of Mean HR and max ΔHR during
the different phases of the seizure.Daggers (*) indicate statistically
different groups .
Figure 5HR average profile aligned according to the seizures onset (black vertical dashed line).
Dotted blue lines delimits the 25%and 75%percentiles.
Figure 4HR slopes during the different phases.The light green area
marks the analysis related to the ictal phase.
* *
... The false alarm rate was lower than 0.4% and the sensitivity over 97% [26]. In another study by empatica's researches [8], it was explored how to detect non-convulsive seizures using heart rate, blood oxygen and electrodermal activity. They found a change in both heart rate and EDA and claim that these sensors might be useful in detecting non-convulsive seizures. ...
ResearchGate has not been able to resolve any references for this publication.