The latest psychology news!
We are one step closer to predicting the unpredictable. Robin Gras, PhD, an associate professor in the School of Computer Science and Canada Research Chair in Learning and Simulation for Theoretical Biology, and his PhD student Abbas Golestanti have developed novel methods for long-term time series forecasting.
In a Scientific Reports article, Gras and Golestanti demonstrate their software’s accuracy with the prediction of earthquakes, financial markets, and epileptic seizures. By applying their algorithm to the EEGs of 21 patients, they were able to predict seizures 17 minutes before the onset with 100% sensitivity and specificity.
Here, I interview Dr. Gras on his prediction software and potential future medical applications.
How does your prediction software work?
Gras: We use a measure of the level of chaos of a time series to predict its future values. For financial or global temperature time series, we have discovered that the level of chaos is very stable, whereas for epileptic seizures, we have discovered that the level of chaos of the EEG time series strongly increases few minutes before the seizure occurs. We use these properties to make our predictions.
Can you summarize your study on epileptic seizures?
Gras: Our new method for complex time series prediction is based on the concepts of chaos theory and an optimization process. The general idea is to extract a unique non-linear characteristic from an existing time series that somehow represents the behavior of the time series and to subsequently generate successive new values that continue the time series, each value minimizing the difference between the characteristic of the new time series and the initial one.
In the case of epileptic seizure prediction, we transform the EEG signal time series into another time series that represents the evolution in time of the level of chaos of the EEG signal. Then, we use our method to predict the future values of the level of chaos of the EEG signal. When this level of chaos become greater than a predetermined threshold, we predict that an epileptic seizure will occur. Our tests on 21 patients data show that this approach can make reliable predictions up to 17 minutes in advance.
What are the benefits of seizure prediction?
Gras: If a patient has a device measuring EEG signal on his head which is connected to a small computer (a smartphone, for example) running our software, the software can create an alarm up to 17 minutes in advance of a seizure giving the possibility to the patient to call a doctor, to stop his car, or to go to a safe place before the seizure occurs.
What other medical applications have you envisioned for your software?
Gras: We would like also to test for heart attack or stroke.
Are you seeking collaborations with the biomedical community?
Gras: We have deposited two patents on this method, and we’re searching for an industrial partner, which could design a sensor device and run clinical tests.
Brain Blogger http://ift.tt/1wKlwc9