Conference PaperPDF Available

Prospective evaluation of seizure detection performance of the Embrace wristband on pediatric patients in the epilepsy monitoring unit.

Authors:
GTCS detection
Embrace detected
12 out of 14 focal to bilateral tonic-clonic seizures in 5 patients
(overall
Sens = 86%
, 95% CI: [70%-88%]; avg. patient-level Sens = 92%).
The
mean detection latency
with respect to the start of the generalized phase was
30 s
(range: 15-49 s).
False alarms
The
overall FAR
was
0.8
(95% CI: [0.48-1.18]).
52% of patients
experienced
0 FA
, while 48% (14 patients) experienced
29 false alarms
.
Reasons for false alerts included patient beating on the tray with her hand, scraping, and
rubbing to disinfect hands.
Seizures other than GTCSs
53 seizures other than GTCSs
were experienced during the recordings: 43 focal motor tonic
from 9 patients, 1 focal automatisms, 1 hyperkinetic, 2 spasms, 1 focal motor clonic, 1 non
motor subjective only, 1 focal motor without impairment, 1 myoclonic.
Among them, 11 focal motor tonic seizures from 4 patients were detected by Embrace.
Prospective evaluation of seizure detection performance of the Embrace
wristband on pediatric patients in the epilepsy monitoring unit
C. Caborni1, F. Onorati1, G. Regalia1, R. Picard1,2,P.De Liso3, L. Fusco3, L. M. Piscitello3, F. Vigevano3
1. Empatica Inc, Boston, MA, US; 2. Massachusetts Institute of Technology, Cambridge, MA, US
3. Bambino Gesú Children’s Hospital, IRCSS, Rome, Italy
Materials and Methods
Results
Conclusions
[1]
Onorati et al. (2017). Multicenter clinical assessment of improved wearable multimodal convulsive seizure detectors.
Epilepsia, 58(11), 18701879. https://doi.org/10.1111/epi.13899
[2]
510(k) Premarket Notification Embrace 12/20/2018. (n.d.). Retrieved July 28,2019, from
https://www.accessdata.fda.gov/scripts/cdrh/cfdocs/cfpmn/pmn.cfm?ID=K181861
References
Francesco Onorati
fo@empatica.com
AIM: To report the results of a phase III prospective trial of Embrace in a pediatric EMU
Eleven subjects were discarded from analysis due to lack of compliance or age lower
than 2 yo.
On the
29 patients
left (8 females,
Table below
), the overall number of days of data
collected was
36.2 days
(868 hours, mean of 30 hours/patient).
No device deficiency or adverse event were recorded.
For the
first time
, the performance of the Embrace
GTCS detection algorithm
was reported on a
prospective EMU pediatric trial
.
Both the Sens and FAR indicate that the
Embrace generalizes well
to motor and autonomic seizures
patterns sensed on the analyzed pediatric cohort, meeting
performance endpoints
previously
accepted by the
FDA
on wrist EDA and ACM data (Sens > 70%, FAR <= 2/day).
Evaluation on seizure data from
another clinical center
involved in this prospective trial is underway.
40 pediatric patients
at risk of having GTCSs were admitted to
the
EMU
at Bambino Gesù Children’s Hospital (Rome) and
monitored with
video-EEG
(vEEG) and Embrace, between
October 2017 and January 2019.
Seizures experienced during the recordings were labeled by
three independent epileptologists
, (following the
ILAE 2016
classification
) who were blinded to sensor data.
Embrace alerts were logged in a separate database.
Sens
was computed as the number of GTCSs detected by the
classifier divided by the total number of GTCS experienced.
FAR
was computed as the overall number of detections not
validated by epileptologists as GTCS, divided by the total of
recorded hours, normalized by 24 hrs.
Patient
ID
Gender
Age Etiology
Video
-
EEG
duration
(hours) GTCS GTCS
detected False Alerts Month
1male 8
Focal
symptomatic (ganglioma) 13 5 3 0 Oct-17
2male 2
Focal
symptomatic (frontal dysplasia) 5 0 0 0 Dec-17
3male 13
Focal
symptomatic (hypothalamic
hamartoma)
27 0 0 0 Jan-18
4male 9
Focal
symptomatic (frontal dysplasia) 77 1 1 2 Jan-18
5female 12
Focal
symptomatic (astrocitoma) 14 0 0 0 Jan-18
7male 7
Focal
symptomatic (parietal dysplasia) 27 0 0 4 Jan-18
8male 22
Focal symptomatic (temporal dysplasia)
47 0 0 1 Jan-18
9male 2
Generalized myoclonic
symptomatic 23 0 0 2 Feb-18
10 female 16
Focal
symptomatic (frontal dysplasia) 21 0 0 0 Mar-18
11 female 17
Focal symptomatic (brain malformation)
7 0 0 0 Mar-18
12 female 5
Focal symptomatic (brain malformation)
4 0 0 0 Mar-18
13 male 3
Focal
symptomatic (
tuberous sclerosis)
24 0 0 0 Apr-18
14 male 2
Focal
symptomatic (STXBP1 mutation) 18 0 0 1 Apr-18
16 female 3
Generalized
symptomatic 24 0 0 2 May-18
17 male 3
Focal idiopathic
24 0 0 0 May-18
18 male 20
Generalized idiopathic
15 0 0 0 Jun-18
20 male 18
Focal
symptomatic (ischemia) 40 0 0 0 Jul-18
22 male 9
Focal
symptomatic 64 3 3 2 Jul-18
23 male 11
Focal
symptomatic 16 3 3 1 Jul-18
24 male 10
Focal
symptomatic (central dysplasia) 43 0 0 2 Aug-18
25 male 12
Focal symptomatic (brain malformation)
39 0 0 5 Aug-18
26 male 8
Focal
symptomatic (precentral
dysplasia)
24 0 0 1 Sep-18
30 male 17
Focal
symptomatic (Rassmussen
encephalitis)
39 0 0 0 Oct-18
32 female 4
Focal
symptomatic (hypothalamic
hamartoma)
42 0 0 0 Oct-18
36 male 2
Focal symptomatic (temporal dysplasia)
81 0 0 3 Nov-18
37 male 13
Focal
symptomatic (perinatal
ischemia)
41 0 0 0 Dec-18
38 female 15
Focal
symptomatic (Noonan syndrome) 8 2 2 0 Dec-18
40 male 12
Focal
symptomatic (complex brain
malformation)
41 0 0 1 Jan-19
41 female 13
Focal
symptomatic (complex brain
malformation)
20 0 0 2 Jan-19
Background
Embrace
(Empatica Inc., Boston, MA) is an FDA cleared wristband that senses
accelerometry
(ACM) and
electrodermal
activity
(EDA) to detect patterns likely associated with
generalized tonic-clonic seizures
(GTCSs) (
Figure 1
).
A
multi-center retrospective study
on 69 EMU patients (4-60 yo) showed the potential of a machine learning algorithm to
recognize GTCSs from wrist-sensor data, with a
sensitivity
(Sens)
higher than 94%
and
false alarm rate
(FAR)
lower than
0.25/day [1]
.
A successive
prospective study
on 141 EMU patients (6-63 yo) showed
Sens of 98%
and
FAR of 0.94
, which led to FDA
clearance of Embrace for ages 6 and up
[2]
.
Higher FAR
occurred for
pediatric patients
, who are more prone to engage in
repetitive movements (e.g. clapping).
Figure 1 Embrace wristband detects an event and transmits an alert to a smartphone, which generates a call via a cloud-based service to designated caregivers.
Figure 2 Wrist accelerometry (ACM) and electrodermal activity (EDA) raw signals of
two GTCSs
experienced by Patient ID 1.
Left
:
detected seizure.
Right
: missed seizure.
Paola De Liso
paola.deliso@opbg.net
Federico Vigevano
federico.vigevano@opbg.net
Contacts
... Different types of seizures and wearable devices for detecting seizures are reviewed in [86]. e study in [63] included the highest number of participants (135) and used Embrace Empatica Watch [64] with an accelerometer and EDA sensors. ...
... In [62], the authors investigated the use of support vector machine (SVM), random forest (RF), naive Bayes (NB), K-nearest neighbor (KNN), and neural network (NN) to diagnose an epileptic seizure based on EEG sampled dataset available at the UCI machine learning repository. Similarly, the authors in [64,65] have used both EDA and accelerometer data for detecting seizures but with different datasets and techniques. ...
Article
Full-text available
Using artificial intelligence and machine learning techniques in healthcare applications has been actively researched over the last few years. It holds promising opportunities as it is used to track human activities and vital signs using wearable devices and assist in diseases’ diagnosis, and it can play a great role in elderly care and patient’s health monitoring and diagnostics. With the great technological advances in medical sensors and miniaturization of electronic chips in the recent five years, more applications are being researched and developed for wearable devices. Despite the remarkable growth of using smart watches and other wearable devices, a few of these massive research efforts for machine learning applications have found their way to market. In this study, a review of the different areas of the recent machine learning research for healthcare wearable devices is presented. Different challenges facing machine learning applications on wearable devices are discussed. Potential solutions from the literature are presented, and areas open for improvement and further research are highlighted.
ResearchGate has not been able to resolve any references for this publication.