SpecParam 2.0: Spectral parameterization with time-resolved estimates and updated model forms

In the analysis of neuro-electrophysiological recordings, recent developments have emphasized that as well as the commonly analyzed rhythmic or oscillatory activity, there is also prominent arrhythmic, or aperiodic activity. The distinction between periodic and aperiodic activity is important – methodologically because these two components can be conflated by common analysis approaches that do not explicitly consider the different forms of the data, and scientifically, as the two features have distinct interpretations in terms of the underlying circuit dynamics and physiological properties. In recent work, we proposed a method for parameterizing periodic and aperiodic components from neural power spectra ( ‘fitting-oscillations and one-over f’ or ‘fooof’), which detects and quantifying frequency-specific peaks of power (putative oscillations) over and above a separately parameterized aperiodic component that contributes power across all frequencies. In this work, we first present an overview of recent updates to this method, that has been generalized and renamed spectral parameterization (‘specparam’), which now supports additional fitting functions that allow for different variants of periodic and aperiodic activity – as well as having improved model selection and evaluations, and in-built capabilities for time-resolved estimations. We demonstrate and evaluate these methodological updates in empirical datasets, including in intracranial recordings whereby the updated models allow for better capturing patterns of aperiodic neural activity, and in which time-resolved analyses are demonstrated to reflect transitions between brain states. Collectively, these analyses demonstrate the utility of explicitly parameterizing aperiodic and periodic neural activity, and in particular support the generalizations made to the method.