Interview with Rosalind Picard—Behind Empatica’s Research
Dear supporters,
Today we want to share with you one of the most interesting webinars hosted so far: the Ask Me Anything with Dr. Rosalind Picard, Empatica’s Chief Scientist and MIT Professor. During the recording of the webinar we were impressed by the quality and quantity of the questions received, so we decided to transcribe the answers and make them available for everyone on our blog. Make yourself comfortable, this is a long post.
Please remember that Embrace is a medical Product & News for seizure monitoring only in Europe so far — the discussion below about seizures pertains to our ongoing scientific work and to the detection algorithm. For users in the United States, that is available only as part of a clinical trial using the Alert app. Yes, you read correctly: Embrace is a medical device in the European Union!
How does Embrace’s seizure detection work?
The Embrace measures three main kinds of data: movement, electrodermal activity (shortened as EDA — represents the electrical changes on the surface of the skin not necessarily involving sweat) and peripheral temperature. You can read more about Embrace specifications here.
Devices on the market usually use two pieces of information: how strong and how long you’re moving. Instead Embrace uses more complex information, combining a lot of aspects of the movement, and completely novel information, electrodermal activity variations that start when regions deep in your brain are activated in an atypical way (as with seizures). Our algorithm uses advanced machine learning methods, and it is trained on data collected in epilepsy monitoring units in hospitals in order to recognize the complex patterns of movements and electrodermal activity that are most likely to accompany a convulsive seizure.
How accurate is the Embrace’s seizure detection algorithm?
We’re committed to presenting our work in top scientific meetings and journals. Most recently we presented results in posters (you can download here) at the Partners Against Mortality in Epilepsy (PAME) Meeting in Virginia and at the Annual Epilepsy Meeting in Prague. We will be presenting new case study results at the American Epilepsy Society meeting in December in Houston.
To test the accuracy of our seizure detection algorithm we use a dataset of hundreds of verified seizures from epilepsy monitoring units. Using this set we can observe different rates of sensitivity, namely the percentage of correctly detected seizures.
One challenge is that the more sensitive a method is to detect true seizures, the higher chance it will also detect 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 PAME meeting we obtained up to 100% sensitivity with an average false alarm rate of less than one a day.
At the Prague meeting, testing on a different set of data, we showed that 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. In other work with new data at Empatica, we are reaching 95% sensitivity with fewer than 2 false alarms over ten days, on average. Results can thus vary a lot: for some people false alarms are rare, while for others they may occur more frequently throughout their day.
Can I adjust Embrace’s sensitivity for my special case?
At this moment it is not possible, but it’s something that we are looking into. Embrace now uses one algorithm for seizure detection and we’re continuously improving it. We’re working to get this algorithm approved by the FDA. What we hope is that several algorithms will be approved, or several tuning parameters for one algorithm, so that everyone could use the most effective sensitivity level for themselves.
Can you improve the accuracy from my data?
We’re continuously working to improve the accuracy of our algorithm while we collect seizure data. When you label a false alarm (if you agreed to participate in the clinical trial) it helps us to improve it! Additionally make sure to label all your seizures with the time that they happen, even if Embrace misses them. Just add them to the diary in the Mate app.
Each time you download the newest version of the app and firmware you will get the most up-to-date and most accurate new version of the algorithm.
What affects Electrodermal Activity (EDA)?
There are lots of things that can lead to high EDA, such as anything that makes you sweat (sports, excitement, stress). We’re analyzing characteristic EDA signatures during seizures so we can decrease the number of false alarms which can be caused by different situations like when people are excited, clapping, jumping up and down, or playing drums.
Electrodermal activity is affected by skin hydration, placement and other biochemical factors inside you. We’ve carefully designed our algorithm to set precedence for values that are invariable and not subject to change as your daily hydration levels vary. While we think it’s good to drink a lot of water and keep your skin moisturized, not everybody does those things, and we still want Embrace to work for everyone.
There are some medications that interact with the electrodermal activity and some people have extremely low electrodermal activity level. Our bodies are very complex, and with so many things changing every day, it can happen that sometimes Embrace will detect convulsive seizures and other times not.
Will the Embrace be able to detect other types of seizures besides tonic-clonic?
The current algorithm is only designed to detect convulsive seizures. It’s hard for us to say exactly what Embrace will or won’t detect in the future, as we need a lot more data and testing first. For now our priority is to focus on accurately detecting tonic-clonic seizures but we would like to develop some studies in other directions.
For instance, we discovered during tests in the Boston Children’s and Brigham and Women’s hospitals that electrodermal activity rises significantly from 42% to 86% of complex partial seizures.
However, since many things make electrodermal activity rise, and complex partial seizures do not have a convulsive motor component, a detector that uses only these data is likely to have a lot of false alarms.
The data measured by Embrace usually does not have a significant response with absence or with psychogenic non-epileptic seizures, although there are some exceptions.
Can the Embrace predict seizures?
We are not able to predict seizures today. Embrace can only detect events that involve both unusual motor activity and electrodermal changes. This is an area we’re extremely interested in. We will let you know if in the future it will be possible but it is something that is going to require much more time, research and lots of data. You can read more: Can Embrace prevent seizures?
On which wrist should I wear my Embrace?
We recommend wearing Embrace on the wrist that generates the most movement during seizures. Movement is an important part of what we measure so for example if you sleep on your right side and both arms tend to move during the seizure we would recommend to put it on your left side.
Try to wear it where you’re less likely to be lying on top of it because it can block also the Bluetooth signal. If both arms are equally likely to move during your convulsive seizures, then start out by putting it on the side that is your non-dominant hand (e.g. put it on your left wrist if you are right-handed.) You can also read this discussion in our community.
What’s next for Empatica’s research team?
We’re working on adding an app to vibrate the Embrace when the EDA goes up to certain levels and it could be very interesting for whom interested in anxiety and stress monitoring. We’re committed to getting that to you, so you can learn more about how your EDA changes with your stress.
We would like to release it by late December, in time for your new year’s resolutions! In the meantime, stay tuned for the release of the Mate app to look at your sleep and physical activity patterns. We welcome your feedback and suggestions on how we can help make Embrace a helpful and pleasing experience for you.
We hope this post has provided a deeper understanding of Embrace’s functionalities and the research behind the device. Should you have any additional questions, please visit our newly launched Community forum and take part in the conversation :)
We would also like to hear from you which topics you’d like us to cover in our future webinars (you can register here!); this will help us meet your expectations and improve the overall webinar experience too.
Thanks so much in advance for your precious cooperation. We look forward to engaging with you all on the Embrace Community forum!
The Empatica Team