Integrating Data Across Oscillatory Power Bands Predicts the Epileptogenic Zone: the Frequency Range Explorer Epileptogenic Zone (FREEZ) Identification Algorithm

Abstract

AbstractEpilepsy affects over 70 million people globally. One-third of people with focal epilepsy have drug-resistant epilepsy. Identification and removal of the site of onset of seizures, termed the epileptogenic zone (EZ), is the most successful treatment to stop seizures in these people. Implanting electrodes into the brain with intracranial electroencephalography (iEEG) is the gold standard diagnostic test for identifying the EZ. But identification of the EZ with iEEG remains challenging in many cases. We developed a novel computational methodology that integrates mean power across delta, theta, alpha, beta, gamma, and high gamma frequencies over time to identify the EZ. A machine learning model was trained to predict electrodes within the EZ using publicly available data from 21 patients. In patients that were seizure free after surgery, electrodes within the EZ had significantly higher area under the curve (AUC) for mean power over time in the first 20 seconds after a seizure compared to electrodes outside the EZ in the alpha (p = 0.001), beta (p = 0.001), gamma (p <0.0001), and high gamma (p = 0.0003) ranges. Additionally, electrodes within the EZ in patients that became seizure-free after surgery had significantly higher AUC compared to electrodes marked within the EZ in patients who did not become seizure-free after surgery in the gamma (p = 0.0006) and high gamma (p <0.0001) power ranges. Leave-one-out patient cross validation of the machine learning model yielded a 95.7% positive predictive value and 80.6% specificity for identifying electrodes within the epileptogenic zone, and 90.5% accuracy for predicting seizure outcome based on a planned resection.We implemented this algorithm into the open-source software tool “Reproducible Analysis and Visualization of iEEG” (RAVE) to enable users to reproduce our results and implement this methodology with new datasets, creating a software module we title FREEZ. The software facilitates quantification of the spectral power changes during seizures, including displaying time-frequency spectrograms and projecting results across patient-specific 3D brain maps. Users can also adjust parameters for visualizing multiple frequency ranges from various time regions around seizure onsets in a web-browser-based interface.