Office of Technology Transfer – University of Michigan

Automated Detector and Classifier of High Frequency Oscillations

Technology #6344

Epilepsy is one of the most frequent neurological diseases with an overall incidence between 0.5 and 1%. The most promising treatment is the surgical removal of the epileptogenic zone (EZ). Patients however can only profit from this type of treatment if seizures are generated over a well-localized area and if this area can be removed safely. There is still a major challenge in treating epilepsy, as the mechanisms of cause are not well understood.

Clinical care still relies upon reading electroencephalograms (EEGs), using techniques that were developed in the 1930’s. Over the past 20 years, researchers have discovered that higher resolution, research grade EEG identifies new signals that were never seen before and have strong correlation with epilepsy. The most well-known of these new biomarkers of epilepsy are High Frequency Oscillations (HFOs), which are oscillations > 80 Hz. HFO detection algorithms have been available for several years, however, the most common detector is highly sensitive, but quite prone to identifying artifacts as HFOs.

Automated detector and classifier of high frequency oscillations

While there have been several strategies to detect HFOs automatically, there are several reasons that make it difficult for them to be used in standard clinical practice. First, each was developed and tuned to a specific dataset and acquisition parameters. Second, in long term EEG there are frequently periods of poor data quality in which automated algorithms are unreliable. Third, any algorithm must account for false positive detections from transient artifacts. The novelty of the strategy in this work is that it accounts for each of these weaknesses to develop a universal detector.

The proposed software can result in two possible determinations: 1) that the given HFO data is not predictive of the seizure onset zone, or 2) that the given HFO data is predictive, and which specific channels are predicted. The invented algorithm accomplishes this by using a combination of thresholds and by examining the distribution of rates. The algorithm combs predictions from overlapping data samples to make a final prediction.

The current invention is an automated method that identifies HFOs and presents the data in a form that clinicians can utilize in their typical workflow. The goal is to locate and display this new biomarker, while providing rigorous data that helps clinicians determine its clinical import. The clinical outcome would be to influence the identification of seizure networks in patients undergoing surgery for epilepsy. HFOs have great potential to identify seizure networks faster and with greater precision than current methods.


  • The HFO information and seizure onset zone detector can be visualized as ‘instruments’ in Persyst’s Magic Marker Tool, thus incorporated into a common clinical EEG software.
  • The proposed software package could be used by researchers analyzing previously recorded data could also could use the HFO detection and distinguishing methods.


  • Real-time detection and visualization of HFOs in clinical EEG viewing software
  • Real-time identification of artifacts
  • Real-time distinguishing of pathological and normal HFOs
  • First method to provide visualization of HFO information to clinicians
  • First prospective algorithm to predict the seizure onset zone
  • Automated, flexible determination of how high of an HFO rate is pathological
  • No algorithms require per-patient tuning: one set of parameters is used for all patients for all algorithms