Canine electrode locations and data segments. (A) For the canine subjects, bilateral pairs of 4-contact strips were implanted oriented along the anterior-posterior direction. Electrode wires were tunnelled through the neck and connected to an implanted telemetry device secured beneath the latissimus dorsi muscle. (B) An hour of data with a 5-min offset before each lead seizure was extracted and split into 10-min segments for analysis. (C) The expanded view illustrates a ∼35-s long seizure.
It might sound like a riddle: What do you get when you combine 1 online contest, 2 patients, 5 five dogs and 654 data scientists? The answer: Hope for patients with epilepsy that their seizures can be reliably predicted, and perhaps prevented. A new study in the journal Brain reports on a Jun, 2014 online contest in which >500 teams of data scientists from all over the world analyzed hundreds of hours of recordings of electrical activity in the brains of 2 people and 5 dogs before and during epileptic seizures.
The contest result? The wisdom of the crowd created a better-than-random prediction of epileptic seizures. In what has long been a highly contentious field of research, with limited reproducibility of results, this study has made all the data and analysis code freely available. For patients, predictability of seizures could be life changing. “If an algorithm could detect subtle changes in the electrical activity of a person’s brain before a seizure occurs, people with epilepsy could take medications only when needed and possibly reclaim the daily activities many of us take for granted,” says Ben Brinkmann, PhD, a data scientist at Mayo Clinic. ~50 million people worldwide have epilepsy.
Development of reliable seizure prediction algorithms requires data on many seizures and non-seizure brain activity as measured by implanted electroencephalography (EEG). But the large quantities of long-duration data needed are difficult to acquire and have not been available. Typically, EEG recordings lasting 7 days or fewer are taken before surgery when a patient’s medications are reduced to encourage seizures. These recordings offer a limited view of brain function under changing conditions.
Even when a working algorithm is created, investigators typically don’t share algorithm code because of concerns related to privacy, cost of sharing large datasets, protecting their research and intellectual property. There is also a lack of incentives to share data. The result is a vast literature that is difficult to compare and impossible to reproduce.
Enter crowdsourcing and data sharing. “In the hope of winning up to $15,000 in prize money and bragging rights in data science circles, hundreds of algorithm developers, most with little or no experience with epilepsy or EEG, worked countless hours to build, test and rebuild algorithms for seizure forecasting,” Dr. Brinkmann says.
The novel data from canines with epilepsy were obtained with the Seizure Advisory System developed by Seattle-based NeuroVista Corp, an early-stage medical device company that folded in 2014. The datasets and source code remain openly available from the Mayo Clinic Systems Electrophysiology Lab and the International Epilepsy Electrophysiology Portal for researchers to continue developing new algorithms for predicting seizures.
Figure 2 Human implanted electrode locations. Implanted electrodes are visible in X-ray CT images coregistered to the space of the patient’s MRI for the two epilepsy patients whose data was used in this competition. (A) Patient 1 had bitemporal 8-contact penetrating depth electrodes implanted along the axes of the left and right hippocampus. (B) Patient 2 had a 3 × 8 subdural electrode grid placed along the axis of the left temporal lobe and frontal lobe strip electrodes. Spheres represent approximate electrode positions due to post-craniotomy brain surface shift in the CT. Electrodes not used in these experiments have been omitted from this illustration.
“These results support our effort to develop the next generation of epilepsy devices designed to continuously monitor brain activity, forecast and prevent seizures,” says neurologist Greg Worrell, M.D., PhD., Mayo Clinic. “These datasets and source code also serve as a benchmark, allowing new algorithms to be compared to each other and to the algorithms developed in this competition.”
Data scientists from across the globe tested algorithms on nearly 350 seizures recorded over more than 1,500 days and the winners agreed to make their computer code freely available. During the 4-month contest, over half of these crowdsourced algorithms performed better than random predictions. The best performing algorithms accurately predicted more than 70% of seizures when tested on unseen portions of the canine data. “The dog EEG data has great bidirectional benefits, both as a comparative model for human epilepsy and also to directly benefit dogs with naturally occurring epilepsy,” says Ned Patterson, Ph.D., D.V.M., Uni of Minnesota.
The effort to forecast seizures continues under a five-year federal grant awarded to the Mayo team in 2015 from the National Institutes of Health. The grant, part of a presidential initiative aimed at revolutionizing the understanding of the human brain, is called Brain Research Through Advancing Innovative Neurotechnologies, or BRAIN Initiative. Working with Medtronic Inc., the Mayo team will test the safety and efficacy of seizure forecasting initially in canines, with human trials to follow. http://newsnetwork.mayoclinic.org/discussion/crowdsourcing-contest-using-data-from-people-dogs-advances-epileptic-seizure-forecasting/
http://brain.oxfordjournals.org/content/early/2016/03/29/brain.aww045
Recent Comments