A Study of Epileptogenic Foci Localization Algorithm Based on Automatic Detection of Comprehensive Feature HFOs and RF-LR

Abstract

Studies have shown that fast ripples of 250–500 Hz in epileptic electroencephalography (EEG) signals are more pathological and closer to the epileptogenic focus itself compared to ripples of 80–250 Hz. However, artifacts of fast ripples and high-frequency oscillations (HFOs) are easily confused and difficult to discriminate, and manual visual screening is both time-consuming and unable to avoid subjectivity. To this end, this paper presents a method for localizing epileptogenic foci based on the automatic detection of integrated feature HFOs and random forest-logistic regression (RF-LR). In this paper, we first extract multivariate features from the preprocessed epileptic EEG signals, and use the random forest algorithm to filter out three features with high importance, based on which, suspicious leads containing HFOs are identified. Then, wavelet time-frequency maps were used for the primary screening of suspected leads to improve the signal calibration efficiency and further localize HFOs in time and frequency. Finally, a logistic regression model was used to automatically classify and identify ripples and fast ripples in HFOs. The results show that the sensitivity, specificity, and accuracy of the model for detecting ripple are 89.37%, 88.26%, and 90.1%, respectively; the sensitivity, specificity, and accuracy for detecting fast ripple are 94.31%, 94.83%, and 93.46%, respectively. Compared with single features, the multivariate features in this paper more comprehensively characterize the complex epileptic EEG signals and provide more accurate information for the localization of epileptogenic foci. The automatic detection algorithm of HFOs proposed in this paper can analyze a large amount of data in a short time and has a good detection performance, which can help clinicians accurately determine the region of epileptogenic foci.