Automated video‐based detection of nocturnal convulsive seizures in a residential care setting

Abstract

People with epilepsy need assistance and are at risk of sudden death when having convulsive seizures (CS). Automated real-time seizure detection systems can help alert caregivers, but wearable sensors are not always tolerated. We determined algorithm settings and investigated detection performance of a video algorithm to detect CS in a residential care setting. The algorithm calculates power in the 2-6 Hz range relative to 0.5-12.5 Hz range in group velocity signals derived from video-sequence optical flow. A detection threshold was found using a training set consisting of video-electroencephalogaphy (EEG) recordings of 72 CS. A test set consisting of 24 full nights of 12 new subjects in residential care and additional recordings of 50 CS selected randomly was used to estimate performance. All data were analyzed retrospectively. The start and end of CS (generalized clonic and tonic-clonic seizures) and other seizures considered desirable to detect (long generalized tonic, hyperkinetic, and other major seizures) were annotated. The detection threshold was set to the value that obtained 97% sensitivity in the training set. Sensitivity, latency, and false detection rate (FDR) per night were calculated in the test set. A seizure was detected when the algorithm output exceeded the threshold continuously for 2 seconds. With the detection threshold determined in the training set, all CS were detected in the test set (100% sensitivity). Latency was ≤10 seconds in 78% of detections. Three/five hyperkinetic and 6/9 other major seizures were detected. Median FDR was 0.78 per night and no false detections occurred in 9/24 nights. Our algorithm could improve safety unobtrusively by automated real-time detection of CS in video registrations, with an acceptable latency and FDR. The algorithm can also detect some other motor seizures requiring assistance.

Under‐reporting of nocturnal seizures using video‐based home monitoring: a case study on the evaluation of the effect of vagal nerve stimulation

“…The system used in our case study was a marker-free video detection system, the type of which has been proven to be accurate for seizures with motor components. These systems can have sensitivities of up to 97%, with a mean false detection rate (FDR) of 0.78 per night for tonic-clonic seizures (Geertsema et al, 2018) and a specificity and sensitivity of above 85% when distinguishing between focal clonic seizures, myoclonic seizures, and random movement in infants (Karayiannis et al, 2006). At the lowest, the sensitivity can be only 57% (Geertsema et al, 2018) and the specificity just 53% (Lu et al, 2013).…”

Seizure detection and mobile health devices in epilepsy: Update and future developments

“…Movement‐based GTCS detection, primarily using accelerometry sensors, is associated with highly variable sensitivity (31%‐95%) and positive predictive value (4%‐60%) across video‐EEG studies, whereas a field study reported even lower sensitivity (14%) . Conversely, 2 other field studies with high sensitivity are reported in this supplement: one using accelerometry showed a median sensitivity of 90% and a false alarm rate of only 0.1/d for GTCS detection, whereas another one using video‐only in a residential care setting reported 100% sensitivity and a median false alarm rate of 0.78 per night . Multimodal seizure detection is currently characterized by various sensing methods and highly variable sensitivity (4%‐100%) and rate of false alarms (0.25‐20 per 8 hours) …”

Movement‐based seizure detection

“…It is the most specific and easiest to recognize sign during these seizures (see the section on clinical trials). The differences between the various systems are mainly determined by the required duration of the rhythmic movements, varying from 2 seconds during video analysis to 10‐16 seconds in current bed sensors (Emfit, Epi‐Care). An interesting phenomenon, the systematically decreasing frequency of the jerks during the clonic phase, has not yet been used in current movement sensors.…”

“…The noncontact sensors “video” and “radar” allow movement detection by analysis of the optic flow signal or of the variation in reflections. The algorithms to extract the movements are comparable to those for accelerometry or bed sensors, because rate and amplitude of the signal variations are the primary detection parameters . Video and radar are the least obtrusive of all sensor types.…”

“…In a residential population (28 patients for a period of 2‐3 months) Arends J (unpublished data) found that no false alarms were generated by the software of a piezoelectric bed sensor when warning after a minimum rhythmic movement duration of 15 seconds. In a sample taken from the same population, Geertsema et al found that video analysis with a minimum rhythmic movement period of 2 seconds was associated with a significant number of false alarms (0.78 per night).…”