Abstracts

Rationale: Transient high frequency oscillations (HFOs) are field potentials that reflect short-term synchronization of neuronal activity. Studies have demonstrated that HFOs (80-500 Hz) are preferentially localized to the brain region generating spontaneous seizures. Visual analysis of HFO’s is time-consuming and susceptible to the interpreter’s level of experience. Recently, multiple methods for automated detection of HFO were reported with varying degree of accuracy. These automated methods allow reliable information about the distribution and quantification of HFO in large data sets. However, translating these automated methods in clinical practice would require some prior skills in signal processing and programming. To overcome this limitation, we propose to evaluate our prototype of a “plug-and-play” interactive graphical user interface (GUI) that entrenched an amalgamation of these automated methods which could be readily applied in clinical practice for mapping and quantifying HFOs with precision.Methods: Six validated methods of detecting HFO’s were incorporated in a Matlab based interactive GUI - a) Root Mean Square (RMS) in conjunction with linear finite impulse response filter(FIR); b) Complex Morlet Wavelet method (CMOR); c) Matching pursuit (MP); d) Bumps modeling technique(Bump); e) Hilbert-Huang Transform(HHT) and f) Tree Machine learning analysis method(D-Tree). Three critical steps were identified towards implementation of our iterative user interface- 1) software utility which is measured by its ability to map and quantify HFO; 2) software usability which is measured by user’s learnability, efficiency of use and subjective satisfaction; 3) visual representation of the output maps. Here we report our preliminary results of developing a software prototype and tentatively evaluating the primary critical step, ie its utility,in quantifying HFO. Sixty minutes of stereo-EEG recordings sampled at 1024 Hz from three adults (each with 20 minutes epoch) were obtained and visual analysis of HFO was considered the ground truth.Results: Snippets of the GUI with channel counts and raw EEG are represented in Figure1. Initially the GUI allowed supervised detection of HFO in a short segment of recording by implementing automated detection algorithms that have high sensitivity but may also have high false detection rate (FDR). Visual analysis enabled the optimization of detection parameters. In the subsequent step, unsupervised fully automated detection with the optimized parameters was applied in the entire data. The outputs maps were displayed as a color coded matrix with frequencies in y-axis, channel counts in x-axis and intensity of color quantifying HFO’s. Random visual analysis at this stage confirmed the accuracy of the automated detection algorithms.Conclusions: An interactive GUI prototype for automated quantification of HFO is developed that hybridized multiple validated detection methods. Our future direction is to implement and improve software usability for wider acceptance in clinical applications of HFO analysis.