Method for Synthetic Generation of LFP Data for Testing of Feature Extraction Algorithms

Heather J. Breidenbach, Virginia Woods, Uisub Shin, Evan M. Dastin-van Rijn, Mahsa Shoaran, Alik S. Widge

Abstract: Recent interest in closed-loop neuromodulation devices has driven development of algorithms capable of real-time biomarker extraction. Synthetic data for tuning algorithmic parameters in various oscillatory cases is a useful tool but must be generated to model realistic neural behavior. We extracted key oscillatory behaviors from rodent LFPs and used this information to create a realistic generation method for synthetic signal production. We then used the generated signals to optimize the feature extraction performance of a real-time feature extraction algorithm. The results of the algorithm testing closely mirrored results from testing on recorded neural LFPs and resembled this real data more closely than a simplistic model of synthetic neural data.