Epilepsy is a neurological disorder in which patients suffer from seizures, or brief periods of excessive electrical activity in the brain. The majority of patients can have their seizures held in check by anti-seizure medication, or, in more extreme cases, brain surgery to remove the area where the seizures originate. For a minority of patients, however, none of these measures are effective, and they suffer from frequent seizures which can be debilitating and cause permanent brain damage.
The Epilepsy Project is a collaboration between CCLS and the Columbia University Medical School's Computational Neurophysiology Laboratory to create an early warning system for epileptic seizures. CLN has implanted microelectrode arrays in the brains of several patients, which produce brainwave readings at a rate of 30,000 times per second. This data, over 30 terabytes of it, is given to the scientists at CCLS, who are working on developing algorithms to predict seizures based on previous EEG readings. They hope that if a reliable prediction algorithm can be developed, it can be used in an implantable device that will predict when a seizure will occur and deliver an appropriate intervention (either medication or a directed electrical shock) to head off the seizure. A device like this would vastly improve the quality of life of many epileptics, particularly those not responsive to medication.
My (very small) part in this was to test a new method for determining whether different channels (the data from different electrodes) influence each other. This is important because seizures often begin in one small area and spread rapidly across the brain. If this spread occurs in set patterns across seizures, it can be used to predict when future ones will occur. These patterns, whoever, are difficult to detect, and that is where computers come in. I worked this summer on using a machine learning algorithm called Support Vector Regression along with a method for calculating influence between two data sources called Granger Causality, to determine whether any two channels in an EEG affect each other.