Automated Detection and Classification of Interictal HFOs to Improve the Identification of Epileptogenic Zones in Preparation for Epilepsy Surgery
Abstract number :
1.030
Submission category :
1. Basic Mechanisms / 1C. Electrophysiology/High frequency oscillations
Year :
2018
Submission ID :
498382
Source :
www.aesnet.org
Presentation date :
12/1/2018 6:00:00 PM
Published date :
Nov 5, 2018, 18:00 PM
Authors :
David Mogul, Illinois Institute of Technology; Tiwalade Sobayo, Illinois Institute of Technology; and Sina Farahmand, Illinois Institute of Technology
Rationale: For the more than 25 million patients worldwide who have drug-resistant epilepsy, surgical resection of brain regions where seizures arise is often the only alternative therapy. However, the identification of this epileptogenic zone (EZ) is often imprecise. Several studies have suggested that localized high frequency oscillations (HFOs) detected during iEEG recording are a way to spatially locate the EZ. Traditionally, HFO detection is carried out through visual inspection of long hours of iEEG recordings. This is a very time-consuming and tedious process. Methods: In this study, an automated method for detecting and classifying HFOs based on noise-assisted, multivariate empirical mode decomposition (NA-MEMD) is proposed to identify the high-rate HFO areas in interictal, multi-channel, iEEG recordings. The NA-MEMD has the potential to outperform Fourier and Wavelet methods because it is adaptive and produces a set of finite narrowband components. NA-MEMD is also less susceptible than filtering to ringing artifacts. Interictal iEEG data recorded from 10 patients, who subsequently underwent epilepsy surgery at the University Hospital of Zurich, were used in this study. Recordings were made using subdural strip, grid and depth electrodes. Data acquisition was performed using a Neuralynx system with a sampling frequency of 2000 Hz and 0.5-1000 Hz band-pass filtering. Pre-processing of iEEG data involved the measurement of bipolar derivations from adjacent channels to eliminate the confounding effects of common reference signal and volume conduction. Bipolar-derivate signals were then divided into consecutive, non-overlapping, 1-s windows. For each window, NA-MEMD was performed on the time series to detect and classify HFO events; namely, fast-ripple (FR), ripple (R), and fast-ripple concurrent with ripple (FRandR). In each class, channels with high-rate HFO activity (as defined by a rate threshold) were selected as its corresponding HFO areas. Results: To investigate the clinical relevance of the detected HFO areas, the resection ratio (RR) of all three classes of HFO were measured for each patient and then compared with his/her post-surgical outcome. The performance analysis of the proposed HFO detection and classification method resulted in sensitivity equal to 91.08% and false discovery rate (FDR) equal to 7.32%. Based on the obtained RR values for different HFO classes, it was found that patients with high RR for detected HFO-FRandR areas were seizure-free while those with low RR had recurrent seizures. Conclusions: These results support the feasibility of using this methodology to provide an automated algorithm that can be used in concordance with SOZs to better delineate the EZ. This method will be further validated across more pre-surgical candidates but it holds the potential to not only improve surgical outcomes at permanently arresting seizure onset but may also reduce the amount of resected brain tissue necessary to achieve this result. Funding: Support provided by NIH R01 NS092760 to DJM