Abstracts

Rationale: Analyzing spectral features measured from electrophysiological recordings is a critical component of epilepsy research, diagnosis, and treatment. Features of interest typically include epilepsy specific events such as pathological high-frequency oscillations (HFOs), as well as measures computed on prespecified oscillation bands. There is still a lack of consensus on which spectral features and analysis methods may be the most productive as potential biomarkers to assist with understanding, predicting, and treating epilepsy. Similarly, other areas of research have often faced inconsistent and/or unclear findings, prompting recent work that has contributed new and updated conceptual understandings, methodological tools, and potential physiological interpretations on the topic of measuring and interpreting spectral features from neuro-electrophysiological recordings. In this project, we seek to relate this work to the topic of epilepsy, by surveying current practice, examining potential pitfalls, and demonstrating novel methods.

Methods: We first used systematic literature searches to empirically estimate which spectral features are examined in the context of epilepsy. We then used these findings to discuss recent methodological developments that may be applicable to epilepsy data. This includes the conceptualization of electrophysiological data as reflecting a combination of periodic (oscillatory) and aperiodic (1/f-like) components. Methodologically, this motivates the need to explicitly isolate neural oscillations, which can be done using methods such as spectral parameterization (formerly "fooof"). We also survey new methods for time-resolved analysis, such as burst detection. These topics are shown with simulated data, as well as with example demonstrations using openly available datasets of extra- and intra-cranial electroencephalography (EEG) data collected from epilepsy patients.

Results: In the literature analysis, we find that as well as epilepsy specific events such as interictal spikes and pathological HFOs, examinations of spectral features typically include measures of the power or connectivity within oscillation bands, such as alpha, beta, gamma, etc. Related to this, we show how methods that aim to measure oscillations without controlling for aperiodic activity can lead to systematic misestimations and demonstrate how spectral parameterization can isolate periodic activity while also measuring the aperiodic component, which is also a potential feature of interest. For transient events, while detection of epileptic spikes and HFOs typically use time-resolved measures we explore how analyses aimed at estimating oscillatory power and/or connectivity could potentially be improved with the use of novel methods that explicitly measure the temporal occurrence, or "burstiness" of oscillations.

Conclusions: Overall, this project serves as a survey of methodological analyses that are used in epilepsy research and seeks to relate this to recent developments from other related fields of investigation, demonstrating ideas and methods that may be useful to apply to epilepsy data.

Funding: None