Summary: If you are testing PineTime, I recommend that you change the Alarm Frequency Min setting from 3 Hz to 2Hz, and reduce the Alarm Ratio Threshold from 57 to 50. This does not affect Garmin watches.
We have been using PineTime in our ‘Production’ OpenSeizureDetector set-up for a few weeks now (after several months of off-line testing).
I had noticed that we were getting fewer false alarms with PineTime than we had been with Garmin, but our false alarm rate does vary depending on Benjamin’s behaviour.
However this morning he head a tonic-clonic seizure which I would have expected OpenSeizureDetector to detect, but it did not alarm, so I have been looking into the reasons.
It seems that the detected frequency of movement was slightly lower than I would have expected, more like 2 Hz rather than 3 Hz, so it was out of the detection range of the default settings. To address this, I have made the following changes to our settings:
- Alarm Frequency Min: Changed to 2 Hz from the default 3 Hz.
- Alarm Ratio Threshold: Changed to 50 from the default 57
I have ran these changes through the database of seizure data, and it does increase the detection reliability as we would expect. It is also giving more false alarms on the Normal Daily Activity database, but I do not know yet how significant this will be in practice – I will update this once we have tried these settings for a bit longer.
In the interim, I recommend that PineTime users change the Alarm Frequency Min setting from 3 Hz to 2Hz, and reduce the Alarm Ratio Threshold from 57 to 50. This does not affect Garmin watches.
I will investigate the PineTime performance more closely to see if it is a sampling rate issue compared to Garmin, so this might involve a change to the PineTime firmware to fix it.
