Unsupervised classification of high-frequency oscillations in neocortical epilepsy and control patients

Quality of life for the more than 15 million people with drug-resistant epilepsy is tied to how precisely the brain areas responsible for generating their seizures can be localized. High-frequency (100-500 Hz) field-potential oscillations (HFOs) are emerging as a candidate biomarker for epileptogenic networks, but quantitative HFO studies are hampered by selection bias arising out of the need to reduce large volumes of data in the absence of capable automated processing methods. In this thesis, I introduce and evaluate an algorithm for the automatic detection and classification of HFOs that can be deployed without human intervention across long, continuous data records from large numbers of patients. I then use the algorithm in analyzing unique macro- and microelectrode intracranial electroencephalographic recordings from human neocortical epilepsy patients and controls. A central finding is that one class of HFOs discovered by the algorithm (median bandpassed spectral centroid ∼140 Hz) is more prevalent in the seizure onset zone than outside. The outcomes of this work add to our understanding of epileptogenic networks and are suitable for near-term translation into improved surgical and device-based treatments.