Crowdsourcing Seizure Prediction

I wrote a short piece for the December issue of Discover Magazine about  freelance coders  recruited by epilepsy researchers to help advance the tricky art of seizure prediction.

For many of the 50 million people with epilepsy, seizures are brutally unpredictable. They can strike out of the clear blue sky at any time; those who suffer them have to be ready all the time. Even people who average only a seizure a month can't drive and shouldn't swim or bicycle alone. Cooking is treacherous. Every sharp corner, high place, or moving vehicle poses risk.  What’s more, to keep seizures at a minimum, many patients suffer the side effects of constant medication. If seizures were predictable, people with epilepsy could conduct business as usual most of the time, getting out of harm’s way when a seizure approaches, or, better still, taking medical action to forestall it.

Over the past 15 years the NIH and several other funding entities have spent about $40 million trying to devise algorithms that detect signs of a brewing neurologic storm early enough to intervene. A driving patient could get off the road. Or a device could deliver anti-seizure medication or a subtle, seizure-aborting electric current.

Seizure prediction proved a tough nut to crack; all that investment produced algorithms little better than a coin flip. Frustrated, a group of neuroscientist epilepsy docs--backed by the NIH, Epilepsy Foundation, and American Epilepsy Society--invited computational geeks and hackers around the world to take a shot. Over 500 teams competed; each was given data sets recorded from the brains of human epilepsy patients and epileptic dogs.

Three months—and $30,000 of prize money--later, winners had produced algorithms predicting seizures with better than 84% accuracy; good enough to one day transform millions of lives.

The new predictive tools are of special interest to three companies already manufacturing implantable stimulation devices that interrupt seizures. Currently, these devices detect seizures but can’t predict them. Adding prediction will allow now tentative patients to seize the day.