Optimization of Automated HFO Detection Parameters Using Visually Marked Events
Abstract number :
3.125
Submission category :
2. Translational Research / 2C. Biomarkers
Year :
2022
Submission ID :
2204816
Source :
www.aesnet.org
Presentation date :
12/5/2022 12:00:00 PM
Published date :
Nov 22, 2022, 05:26 AM
Authors :
Trisha Mendoza, BS – University of California, Irvine; Jack Lin, MD – Professor, Biomedical Engineering, Anatomy & Neurobiology, University of California, Irvine; Indranil Sen-Gupta, MD – University of California, Irvine; Beth Lopour, PhD – Professor, Biomedical Engineering, University of California,Irvine; Casey Trevino, Student – University of California, Irvine
Rationale: Many studies have suggested that high frequency oscillations (HFOs) are a biomarker of the seizure onset zone (SOZ) and a potential tool for surgical planning. However, despite strong results on the group level, removal of HFO-generating regions has not been found to be predictive of individual surgical outcomes. One possible source of this variation is the method of HFO detection. The development of automated methods has led to fast and objective HFO identification, but there are no accepted guidelines for parameter selection (e.g., the threshold). Visual detection could provide a foundation to determine the best parameter settings for each patient. To test this idea, we optimized automatic HFO detection parameters in each subject using visually detected events and compared the accuracy of SOZ localization using these optimal parameters to the accuracy when using standard HFO detection parameters.
Methods: A total of 19 patients diagnosed with refractory epilepsy undergoing surgical evaluation were included in this study, and all had a postoperative outcome of Engel Class I. Five 3-minute iEEG segments were randomly selected for HFO detection using the RMS detector [1]. To determine the impact of visual HFO detection on parameter selection, we visually marked HFOs in one minute of iEEG data for each patient. We then compared the visually-detected events to those obtained via automatic detection using a range of different parameters, and we selected the parameter set with the highest F-score for each patient. We then used those optimal parameters to detect HFOs in the five iEEG data segments and computed the F-score for classification of SOZ and non-SOZ channels for each segment. We compared these results to the F-scores using standard detection parameters from the original publication [1], using a Wilcoxon rank sum test for each subject, with the Benjamini Hochberg procedure to correct for false discovery rate.
Results: Of the 19 patients analyzed, 5 patients had F-scores that were significantly different for the optimized parameters compared to the standard parameters (p< 0.05). Out of the 5 patients, 4 patients had significantly higher F-scores when visual detection was used to optimize automated HFO detection, while 1 patient had a significantly lower f-score when visual detection was used. We also tested optimizing the detection parameters for each channel (rather than each subject), but this method produced qualitatively similar results.
Translational Research