Contents
Real Tonic-Clonic Seizures
14 November 2019
Benjamin had a tonic clonic seizure just before 06:00 on 14 November 2019. It was a relatively small one from which he recovered quickly. He was stiff and lying on his back, propped up on his right arm. He was wearing the Garmin VivoactiveHR watch running the Garmin Seizure Detector V0.7.
The seizure detector settings were an acceleration threshold of 100, an ROI Ratio threshold of 54, with the region of interest defined as 3 to 8 Hz (alarm Frequency Min and Alarm Frequency Max settings)
He had suffered eczema on his writs wearing the watch directly on his wrist, so the watch was on the outside of his clothing. This meant there is no heart rate data for the incident, but we did have the system set to log data , so we have full accelerometer data available.
The data is summarised in the figure below:
The top figure shows that he was in a deep sleep immediately before the seizure (no movement detecte dby the accelerometer. When the region of interest (roi) power increased significantly more than the whole spectrum power (middle graph), which would be expected (because we tuned the detection for this sort of seizure)
The bottom graph shows the ROI Ratio along with the threshold value of 54. The dots show when the system went into the WARNING state (smaller dots) and ALARM (large dot) becuse the ROI ratio exceeded the threshold value for long enough.
Conclusion – the system worked as intended – it gave a warning followed by a full alarm initiation in response to the shaking associated with the seizure.
Other Seizures
29 November 2019
Benjamin had one of his less severe seizures on 29 November as he was waking up. We noticed him propped up at an odd angle on the bed and when we went to see him we found he was having a seizure with his left arm and right leg ‘waving’ in a gentle motion.
The OpenSeizureDetector system did not detect that movement, but as he started to come out of the seizure, he sat up and the system generated a warning once while he was sitting.
Graphs of the data collected below:
The spectra in the third graph show the issue – The 06:12:00 and 06:12:30 spectra show a peak in the 2-3 Hz region, but we have the system set to detect movement in the 3-8 Hz region, so the movement was too slow to generate an alarm. The 06:12:51 and 06:12:56 data points have peaks in 4-5 Hz, which is what generated the warning condition.
Conclusion: This seizure was not the sort that we have the system tuned for – if we wanted to alarm for this condition we would have to move the AlarmFrequencyMin setting down to 2Hz, and I fear this would generate an excess of false alarms – one day I will get around to writing the code to make it easy to test this now I am generating huge amounts of ‘normal’ data with the Garmin watch.