A case-study on learning from large-scale intracranial EEG data using multi-core machines and clusters

Epilepsy is a chronic neurological disorder characterized by recurrent, unprovoked seizures that manifest in a variety of ways, including emotional or behavioral disturbances, convulsive movements, and loss of awareness. The problem of prediction of epileptic seizures is hard and most algorithms do not perform better than a random predictor [20]. An important reason why studies so far have been less than successful is that electroencephalogram (EEG) is not recorded at the granularity of the seizure generation process. Our collaborators at the Columbia University Medical School (CUMC) have been involved in a clinical trial which entails implanting a Micro-Electrode Array directly into the neo-cortex of epilepsy patients undergoing surgery to remove the portion of the brain from where seizures originate. The 96 contact grid allows researchers to record at 30 KHz/channel which is a very high resolution data collection procedure compared to known state-of-the-art techniques and yields both local field and action potential data (.5 TB per patient per day). This large volume of data poses challenges for knowledge discovery and mining.