Classifying high-frequency oscillations in physiological recordings

Method for automatic and unsupervised classification of high-frequency oscillations in physiological recordings.

Problem:

Automatic detection and classification of high frequency oscillations (HFO) data for use in clinical diagnosis and therapy has been limited. Currently, researchers visually inspect for high frequency oscillations.

Solution:

An algorithm to automatically extract quantitative local seizure information from multielectrode data. The automated HFO detection and classification algorithm is comprised of three major stages:
-  In the first stage, candidate HFO events are detected in the band-pass filtered iEEG.
-  In the second stage, a statistical model of the local background iEEG surrounding each candidate event is built. Events bearing too large a spectral similarity to the background activity according to the model are discarded from candidacy.
-  In the final stage, computational features are extracted from the retained candidates and these features are used, after a dimensionality reduction step, as inputs to a classifier.

Advantages:

Algorithm may process brain oscillations within the 100-500 Hz frequency range, action potentials and epileptic spikes, and other brain/body generated signals
Algorithm may be implemented in intracranial electroencephalographic monitoring systems or brain-implantable devices
Algorithm may enable quantitative seizure localization in devices

Applications:

Intracranial electroencephalographic (iEEG) monitoring systems used in surgical evaluation for epilepsy patients who are unresponsive to antiepileptic drugs
Brain-implantable devices that monitor, predict, abort, and/or control epileptic seizures by using the feedback to deliver a therapy

Stage of Development: