An artificial intelligence/machine learning approach to predicting anti-seizure medication efficacy in Dravet syndrome

A major challenge in the management of many epilepsies, including Dravet syndrome (DS), is the prolonged trial and error approach of determining whether or not a specific antiseizure medication (ASM) will work for a given patient. Testing an add-on therapy can take many months before it is determined to be ineffective, and for some it takes years (if ever) to find an effective medication. Artificial intelligence (AI) now offers the potential to remedy this problem.

Dr. Ivan Soltesz’s laboratory at Stanford published a paper this year (see below) based upon an AI/machine learning method they developed and call Motion Sequencing (MoSeq). They applied MoSeq to both acquired (chemoconvulsant-induced) and genetic (SCN1B) mouse models of epilepsy. MoSeq involves sophisticated analysis of 3D video imaging of mouse behavior. After videotaping epileptic and intact mice, MoSeq was able to learn to distinguish the two apart just based upon types of movement without waiting for seizures to occur. Importantly, MoSeq was also able to distinguish mice treated with an ASM that controlled the seizures versus an ASM that did not control seizures, again just based on mouse movements on the medications without having to wait to see the effects on the seizures themselves. Drs. Jack Parent (University of Michigan) and Ivan Soltesz now seek to apply this AI approach to children with DS. They aim to determine whether MoSeq can learn to predict whether or not a child with DS will respond to an ASM based upon videotaping how the medication affects their movements and behavior.

You can help researchers test this technique by submitting a video of your child with Dravet syndrome behaving typically.

To determine whether a pilot study is feasible, they need to apply MoSeq to some sample videotapes of children with DS.

They are not looking for videos of seizures, but rather of children behaving typically in their environment.

The longer the video, the better.

They will also need to determine what ages of the children would be optimal, but we are thinking possibly between ages 2 and 9 years.

An eventual ideal scenario may be 15 or 20 minutes of video shot in a featureless room or in front of a green screen, but for this first step they will work with whatever videos are available.

To participate, please submit videos at this link.

Videos will be stored on ShareFile. ShareFile uses advanced encryption methods when sending and storing your files, which is much more secure than email attachments and standard cloud storage. By submitting a video(s) you are agreeing that they may be shared with the study coordinators. Only DSF staff and the study coordinators will have access to view the videos. 

Questions on the project may be emailed to our Scientific Director, Dr. Veronica Hood.  

This work is based upon the following published mouse study by the Soltesz laboratory:
T. Gschwind et al, Hidden behavioral fingerprints in epilepsy. Neuron. 2023 May 3;111(9):1440-1452.e5. doi: 10.1016/j.neuron.2023.02.003. Epub 2023 Feb 24.

Scroll to Top
Skip to content