HFOApp: A MATLAB Graphical User Interface for High-Frequency Oscillation Marking

Example usage and output of HFOApp. A–C, To visually detect HFOs in our test dataset, we opened the following three windows: a main window that shows raw data (A); a bandpass-filtered window (80–250 Hz for ripple detection; B); and a spectrogram window for the first channel (C). The cursor is indicated as a blue (A, B) or black (C) vertical line. The red dashed rectangle indicates the boundaries of a detected HFO. A summary of the characteristics of the detected event is shown above the plot in blue font.

Multiple-channel HFO detection. When the “Enable multiple channel detection” option is enabled, all events for all channels in the current time widow can be detected by pressing “d” on the keyboard. A, B, Illustration of the automatically synchronized main (raw) and bandpass-filtered windows of the toolbox. The blue vertical line indicates the cursor location, and the red dashed rectangle indicates detected HFOs. C, Example of formatted output of HFOApp. Each row shows detailed information about each event. The first 10 events are shown. D, A summary of events for each channel.

Features of the time series graphical user interface of HFOApp. The main window and bandpass-filter window share the same graphical user interface.

Features of the spectrogram graphical user interface of HFOApp. The layout of the spectrogram window is similar to the main window, with an extra panel added for spectrogram event detection.

Schematic plot of Hilbert detector implementation. A, An example of raw iEEG data containing HFOs. The vertical gray line indicates cursor location. B, Bandpass filtered data (blue) and its envelope (red), obtained using the Hilbert method. C, HFO detection. The envelope time series (B) is first z score normalized. Clusters surviving an initial threshold (onset threshold) within a distance (search range) of the cursor location are identified. The cluster will be considered an HFO event if the following two criteria are met: (1) the maximal z score is greater than a user-specified threshold (in this example 5); and (2) the number of cycles meets the user-specified value (in this example 3). D, Characteristics of an event. For a detected event (red dashed rectangle), the following characteristics are retrieved: peak z score (blue dot), duration, average frequency, and number of cycles. To estimate the average frequency, oscillatory peaks for the event are identified (red dots) and all peak-to-peak distances are calculated, then the sampling rate is divided by the average peak-to-peak distance. The number of cycles is estimated by dividing the duration of the event by the average peak-to-peak distance. The red dashed boxes in B–D indicate an HFO event.

HFO detection using spectrogram. A, In the spectrogram detection mode, the user clicks on the hotspot of the spectrogram. The black vertical line indicates the location of the click. B, The toolbox computes the peak frequency (red dot) of the spectrogram at the cursor location (A). The HFO is then identified using the Hilbert detector with a narrow bandwidth, which can be specified by the user, centered at the peak frequency. C, A screenshot of the spectrogram window of the toolbox. The black vertical line indicates the cursor location, and the red dashed rectangle indicates the detected HFO.

Validation of HFOApp. A, Proportion of simulated HFOs that were detected by HFOApp and RIPPLELAB default automatic Hilbert detectors. B, Proportion of detected HFOs that were not real simulated HFOs. C, Average duration of detected HFO relative to that of the simulated HFO. D, Computation time. E, Manual marking time for 10 real datasets. F, Inter-rater reliability (Fleiss’ κ) across four raters for the real datasets. The error bars in A–D indicate the SE of the average across 10 simulated datasets. The error bars in E and F indicate the SE of the average across 10 patients. The comparison of time and Fleiss’ κ values between HFOApp and RIPPLELAB was performed using a two-tailed paired t test.