Empatica is constantly working on validating and improving its algorithm. In this article we’ll discuss how we test for accuracy and our latest findings.
To test the accuracy of our seizure detection algorithm we collaborated with top hospitals to collect clinically labelled seizure data using EEG (v-EEG), consisting of 192 recordings taken from 53 patients wearing an Empatica E3 or E4 wrist sensor recording EDA and accelerometer data.
The data were analyzed off-line using proprietary software (Empatica, Inc.). The algorithm tested achieved a seizure detection rate for GTC of 100% on 12 new seizures, while maintaining a false alarm rate of approximately one per day on average over 164 days. We presented these findings at the Partners Against Mortality in Epilepsy (PAME) Meeting in Virginia on June 23-26, 2016.
For the Annual Epilepsy Meeting in Prague on September 11-15, 2016, we presented further findings using a different data set. We were able to reach 90% sensitivity of convulsive seizure detection with an average of 1 false alarm/day, and 95% sensitivity with an average of 2 false alarms a day. When we pushed the sensitivity higher (to 100%) then, on average, there were 5.7 false alarms/day. This shows that results are subject to variation; for some people false alarms are rare, while for others they may occur more frequently.
One challenge is that the more sensitive a method is to detect true seizures, the higher chance it will also produce false alarms (things that look like a seizure to the sensors, but aren’t). Thus, the higher range of sensitivity we set, the more false alarms we have to be willing to tolerate.
In the future, the model will be tested on data collected outside the clinic, where the test conditions are expected to be much more challenging.