Prediction of seizure outcome improved by fast ripples detected in low-noise intraoperative corticogram

•Low-noise EEG increases the signal-to-noise ratio in the fast ripple frequency band.•Unsupervised fast ripples detection standardizes the definition of clinically relevant fast ripple.•Prediction of seizure outcome was improved by the optimal integration of the low-noise EEG and the unsupervised fast ripples detection.

ObjectiveFast ripples (FR, 250-500 Hz) in the intraoperative corticogram have recently been proposed as specific predictors of surgical outcome in epilepsy patients. However, online FR detection is restricted by their low signal-to-noise ratio. Here we propose the integration of low-noise EEG with unsupervised FR detection.MethodsPre- and post-resection ECoG (N = 9 patients) was simultaneously recorded by a commercial device (CD) and by a custom-made low-noise amplifier (LNA). FR were analyzed by an automated detector previously validated on visual markings in a different dataset.ResultsAcross all recordings, in the FR band the background noise was lower in LNA than in CD (p < 0.001). FR rates were higher in LNA than CD recordings (0.9 ± 1.4 vs 0.4 ± 0.9, p < 0.001). Comparison between FR rates in post-resection ECoG and surgery outcome resulted in positive predictive value PPV = 100% in CD and LNA, and negative predictive value NPV = 38% in CD and NPV = 50% for LNA. Prediction accuracy was 44% for CD and 67% for LNA.ConclusionsPrediction of seizure outcome was improved by the optimal integration of low-noise EEG and unsupervised FR detection.SignificanceAccurate, automated and fast FR rating is essential for consideration of FR in the intraoperative setting.