A new localization method for epileptic seizure onset zones based on time-frequency and clustering analysis

• Published in: Pattern recognition 2021-03, Vol.111, p.107687, Article 107687
• Main Authors: Wu, Min, Wan, Ting, Wan, Xiongbo, Fang, Zelin, Du, Yuxiao
• Format: Article
• Language: English
• Subjects: Clustering analysis; Epilepsy; High-frequency oscillations; Seizure onset zones; Time-frequency analysis
• ISSN: 0031-3203; 1873-5142
• DOI: 10.1016/j.patcog.2020.107687

•The Stockwell entropy based on Hilbert transform detects events of interest (EoIs) effectively compared with simple Hilbert transform by accurately detecting both the EoI and non-EoI.•The power method based on Shannon-entropy-based complex Morlet wavelet transform obtains channels of interests with lower computational complexity than the power spectral density method based on SECMWT does.•The adaptive-genetic-algorithm-based matching pursuit (AGA-MP) integrated with the k-medoids clustering method is found to detect high-frequency oscillations (HFOs) more effectively than the AGA-MP method by discerning HFOs from normal activity and artifacts.•The devised new localization method has superiority in improving the localization performance (i.e. sensitivity and specificity) over some existing methods. High-frequency oscillations (HFOs) are spontaneous electroencephalogram patterns that have been regarded as potential biomarkers of epileptic seizure onset zones (SOZs). Accurately detected HFOs are used to localize SOZs, which is crucial for the presurgical assessment. Since the visual marking of HFOs is time-consuming, a method is desirable to automatically detect HFOs for localizing SOZs in clinical practice. However, the existing methods cannot obtain satisfactory performance, which are not suitable for clinical application. In order to solve this problem, we present a new localization method for epileptic SOZs in this study. Firstly, a threshold method is used to detect events of interest (EoIs). Secondly, a time-frequency analysis method is adopted to acquire channels of interest (CoIs) by calculating the average power of EoIs on each channel. Then, the k-medoids clustering method is employed to detect HFOs of CoIs. Finally, the concentrations of detected HFOs are used to localize SOZs. The superiority of our localization method is demonstrated by comparing its sensitivity and specificity with some existing methods.