Abstract
Interictal High-frequency Oscillation (HFO) between 80-600 Hz in intracranial EEG (iEEG) is a promising biomarker of the epileptogenic zone in individuals with epilepsy. Numerous studies revealed that the resection of channels with a high rate of HFOs correlates with favorable surgical outcomes. Early feedback to clinicians regarding the distribution of HFOs during the iEEG recording, especially after the implantation of electrodes, would be helpful for clinical decisions. However, iEEG recording can easily get corrupted by various factors mimicking real HFOs. This study presents a real-time HFO detection framework within MATLAB/Simulink that exhibits robustness against such pseudo-HFOs. This detector is responsible for identifying the initial pool of candidate HFOs and transmitting them via user datagram protocol (UDP) to an external application. The external application contains a machine learning tool that is utilized for post-processing and isolating the real-HFOs. It is implemented in a graphical user interface (GUI) that provides visual feedback regarding the distribution and waveforms of HFOs. The entire processing pipeline was validated by randomly selecting 10-minute segments of interictal iEEG recordings from 10 subjects. It successfully identifies the seizure onset zone (SOZ) in these subjects, achieving an average accuracy of 65% using the detected Ripples and 74% using the detected events with both Ripples and Fast Ripples. Importantly, the spatio-temporal distribution of detected HFOs in real-time showed more than 98% spatio-temporal similarity index compared to offline analysis. Our framework proves to be an effective tool for the automatic identification of HFOs in real-time with the ability to promptly stream the HFO analysis results and provide early feedback regarding the probable SOZ regions to clinicians for surgical decision-making.