0437 Characterizing the Impact of EEG Referencing on Sleep Spindle and Slow Oscillation Analyses

Introduction

The impact of EEG referencing on sleep oscillations, such as spindles and slow oscillations, is largely overlooked across studies. While it is recognized that a topographic head plot of EEG activity does not reflect the true location of the underlying cortical activity, spatial distributions, as well as spectral properties and morphology of EEG oscillations can change dramatically as a function of referencing scheme. It is therefore vital to understand the impact of referencing when drawing inferences about the nature of EEG sleep oscillations. In this study, we use MRI structural data to construct subject-specific forward models of EEG signals. Using these models, we can simulate cortical activity and observe its true representation on the scalp. In particular, we simulate spindles and slow wave oscillations and examine how referencing affects topography, spectral power, and phase of oscillations.

Methods

High-density EEG (Brain Vision, 64-channel) polysomnography was performed on 9 healthy young subjects. 3T structural MRI scans were acquired and forward models were built in MNE-Python using 3-shell Boundary Element Models (BEM) based on individual anatomical details processed with Freesurfer. Simulations of various sleep spindle and slow oscillation dynamics were projected to the sensor space. Different referencing schemes (common average, Laplacian, linked-mastoid) were then applied to the experimental and simulated data and analyzed for effects on time-frequency characteristics of sleep oscillations.

Results

Analyses of experimental data showed distinct reference-based differences in topographical distribution of spectral power and phase of oscillations. Simulated data revealed many scenarios in which the spatial distribution of activity the EEG sensor space poorly represented the true location of the underlying source activity. Moreover, there were alterations to the spatial spread and envelope form of sleep spindle events under different referencing schemes despite from identical source activities.

Conclusion

This study shows that spindle and slow oscillation activity is highly variable across referencing schemes and that EEG topographical plots on the scalp may poorly represent cortical activity locations. It is thus vital to consider the choice of referencing when quantifying characteristics of sleep EEG oscillations.

Support

This work was supported by R01 NS-096177.

0449 Characterizing Spindle Activity as a Time-Frequency Phenomenon

Introduction

Spindles are currently defined clinically based on observed patterns in the EEG waveform trace, with automated methods seeking to replicate visual scoring by experts. Recent work suggests that sleep spindles may be more readily observed as time-frequency peaks in the EEG spectrogram. This study compares spectral peaks in the multitaper spectrogram to expert and automatic detection scoring, characterizes the variability of spindles across a night, and investigates topographical and temporal clustering of spindles within individual EEG records.

Methods

We compared spectral peaks, expert scoring, and automatic detection in two datasets (DREAMS, and a high-density control study). Peaks were identified using multitaper spectral estimation and the peak prominence of the normalized power spectrum for each channel. Spatiotemporal variability analysis was performed using cluster and pattern recognition algorithms including penalized sorting of channel activation order, 2D-cross correlation, PCA and UMAP cluster analysis, and the seqNMF method.

Results

Spectral peaks were shown to be highly robust to and easily differentiated from broadband noise, occuring at rates (10-16 per min) far exceeding spindle rates reported in literature (~2.5 per min). Expert scoring and automated scoring failed to capture clear spectral peaks in the time-frequency domain, indicating an underreporting of the phenomenology. No apparent clustering or patterns of sleep spindle-like activity was observed using the proposed methods, suggesting high variability of spatiotemporal evolution of spindles.

Conclusion

These results suggest that the difficulty of time-domain visual scoring of spindles causes an artificially low estimate of the underlying phenomenology, which is mirrored in the assumptions implicit in the thresholds of automated scorers. This work shows that spindles are highly variable in their spatiotemporal evolution, suggesting that there is no optimal single electrode for analysis and casting doubt on the presence of a single cortical generation mechanism. We must therefore revisit the concept of the spindle using the time-frequency domain to more robustly characterize underlying phenomenology.

Support

National Institute Of Neurological Disorders And Stroke Grant R01 NS-096177

Beamforming approach to phase-amplitude modulation analysis of multi-channel EEG

Phase-amplitude modulation is a form of cross frequency coupling where the phase of one frequency influences the amplitude of another higher frequency. It has been observed in neurophysiological recordings during sensory, motor, and cognitive tasks, as well as during general anesthesia. In this paper, we describe a novel beamforming procedure to improve estimation of phase-amplitude modulation. We apply this method to 64-channel EEG data recorded during propofol general anesthesia. The method improves the sensitivity of phase-amplitude analyses, and can be applied to a variety of multi-channel neuroscience data where phase-amplitude modulation is present.

Characterizing learning by simultaneous analysis of continuous and binary measures of performance

Continuous observations, such as reaction and run times, and binary observations, such as correct/incorrect responses, are recorded routinely in behavioral learning experiments. Although both types of performance measures are often recorded simultaneously, the two have not been used in combination to evaluate learning. We present a state-space model of learning in which the observation process has simultaneously recorded continuous and binary measures of performance. We use these performance measures simultaneously to estimate the model parameters and the unobserved cognitive state process by maximum likelihood using an approximate expectation maximization (EM) algorithm. We introduce the concept of a reaction-time curve and reformulate our previous definitions of the learning curve, the ideal observer curve, the learning trial and between-trial comparisons of performance in terms of the new model. We illustrate the properties of the new model in an analysis of a simulated learning experiment. In the simulated data analysis, simultaneous use of the two measures of performance provided more credible and accurate estimates of the learning than either measure analyzed separately. We also analyze two actual learning experiments in which the performance of rats and of monkeys was tracked across trials by simultaneously recorded reaction and run times and the correct and incorrect responses. In the analysis of the actual experiments, our algorithm gave a straightforward, efficient way to characterize learning by combining continuous and binary measures of performance. This analysis paradigm has implications for characterizing learning and for the more general problem of combining different data types to characterize the properties of a neural system.

A mixed filter algorithm for cognitive state estimation from simultaneously recorded continuous and binary measures of performance

Continuous (reaction times) and binary (correct/ incorrect responses) measures of performance are routinely recorded to track the dynamics of a subject's cognitive state during a learning experiment. Current analyses of experimental data from learning studies do not consider the two performance measures together and do not use the concept of the cognitive state formally to design statistical methods. We develop a mixed filter algorithm to estimate the cognitive state modeled as a linear stochastic dynamical system from simultaneously recorded continuous and binary measures of performance. The mixed filter algorithm has the Kalman filter and the more recently developed recursive filtering algorithm for binary processes as special cases. In the analysis of a simulated learning experiment the mixed filter algorithm provided a more accurate and precise estimate of the cognitive state process than either the Kalman or binary filter alone. In the analysis of an actual learning experiment in which a monkey's performance was tracked by its series of reaction times, and correct and incorrect responses, the mixed filter gave a more complete description of the learning process than either the Kalman or binary filter. These results establish the feasibility of estimating cognitive state from simultaneously recorded continuous and binary performance measures and suggest a way to make practical use of concepts from learning theory in the design of statistical methods for the analysis of data from learning experiments.