The investigators have developed flexible, active, multiplexed recording devices to enable interface with thousands of electrodes implanted on the surface of the brain. While this technology has enabled a much finer view of the electrical activity of the brain, the analytical methods to process, categorize and respond to the huge volumes of data produced by these devices are presently lacking. Many existing neurological data analyses rely on manual inspection. With new neural interfaces with thousands of channels, the data volume is infeasible for manual review. Further, manual inspection can miss subtle features that automated machine learning techniques can detect. In this research, the investigators develop efficient and sensitive automated methods to analyze micro-electrocorticographic (µECoG) data from patients with epilepsy. These methods are used to segment, categorize and predict spatiotemporal epileptiform discharge (or spike) patterns. Understanding the ordering and relationships between these patterns is a key to developing better seizure detection and prediction techniques and ultimately better therapies for patients with epilepsy.
This research comprises four interconnected components. The first component develops techniques for detecting and isolating spike segments, and for extracting features that capture the spatio-temporal pattern of each spike. The second component develops unsupervised clustering algorithms that can identify distinct clusters of spike motion patterns based on carefully chosen features. The thir-d component develops classifiers that can categorize each spike into a few classes (inter-ical, pre-ictal, ictal and post-ictal) based on not only its spatio-temporal pattern, but also the patterns of past spikes. The final component develops methods to predict spike wavefront locations. The combination of these methods will enable seizure prediction and real-time responsive brain stimulation to suppress seizures.
