NeuroDSP: A package for neural digital signal processing

Populations of neurons exhibit time-varying fluctuations in their aggregate activity. These data are often collected using common magneto- and electrophysiological methods, such as magneto or electroencephalography (M/EEG), intracranial EEG (iEEG) or electrocorticography (ECoG), and local field potential (LFP) recordings (Buzsáki, Anastassiou, &Koch, 2012). While there are existing Python tools for digital signal processing (DSP),such as scipy.signal, neural data exhibit specific properties that warrant specialized analysis tools focused on idiosyncrasies of neural data. Features of interest in neural data include periodic properties—such as band-limited oscillations (Buzsáki & Draguhn, 2004)and transient or ‘bursty’ events—as well as an aperiodic signal that is variously referred to as the 1/f-like background (Freeman & Zhai, 2009; Miller, Sorensen, Ojemann, & Nijs,2009), or noise (Voytek et al., 2015), or scale-free activity (He, 2014), and that may carry information about the current generators, such as the ratio of excitation and inhibition(Gao, Peterson, & Voytek, 2017). NeuroDSP is a package specifically designed to be used by neuroscientists for analyzing neural time series data, in particular for examing their time-varying properties related to oscillatory and 1/f-like components.