Poster No:
1653
Submission Type:
Abstract Submission
Authors:
Hayyan Liaqat1, Vasily Vakorin2, Sam Doesburg2, Sylvain Moreno2
Institutions:
1Simon Fraser University, Burnaby, British Columbia, 2Simon Fraser University, Vancouver, British Columbia
First Author:
Co-Author(s):
Sam Doesburg
Simon Fraser University
Vancouver, British Columbia
Introduction:
Neurophysiological parameters characterizing brain organization at various levels of functional and structural hierarchies are commonly observed as following skewed distributions (Buzsáki & Mizuseki, 2014; Roberts et al., 2015). Skewed distributions can take many shapes, but in general, resemble a logarithmic-normal (log-normal) distribution (Koch, 1966). The log-normal distribution has been used to model firing rates of individual neurons (Shafi et al., 2007), spike transmission probability (Mizuseki & Buzsáki, 2013), axon diameters, conduction velocity (Wang et al., 2008). While studies commonly assume that the logarithmic-normal distribution is a suitable model for describing brain parameters, we still have a limited understanding of how accurate this assumption is (Buzsáki & Mizuseki, 2014). In our study, we aimed to compare probability distribution models to determine what models can accurately describe the temporal variability of MEG signals. This can reveal the specific brain mechanisms responsible for the observed variability and guide the use of more accurate statistical tools to compare experimental groups.
Methods:
We analyzed MEG data from the Cambridge Centre for Ageing and Neuroscience (Cam-CAN) Stage 2 cohort study (Shafto et al., 2014). The Cam-CAN study is a population-based, cross-sectional investigation that spans the adult lifespan (18–89 years old). We included 646 healthy adults, dividing them into three age groups: younger (18-45 years old), middle-aged (45-66 years old), and older (66-89 years old) adults, with males and females considered separately.
We analyzed the variability of frequency-specific MEG power fluctuations at five frequencies: 2 Hz, 6 Hz, 12 Hz, 24 Hz, and 48 Hz. For each participant and frequency, we randomly selected a MEG channel from the MEG gradiometers and then randomly selected a 30-second segment. We then modeled the variability of MEG signals power across time by fitting their corresponding empirical distributions to a large pool of theoretical probability distribution models with a maximum likelihood criterion. In total, we evaluated the goodness of fit for 72 statistical models applied to each empirical distribution. We then ranked the best-performing models. To distinguish the performance of each theoretical distribution model, we performed a pairwise comparison of the goodness of fit between each pair of models.
Results:
Our results revealed that two models, the Generalized Gamma and Exponentiated Weibull distributions, provided the best fit of the temporal variability of MEG signals, with undistinguishable performance irrespective of age, sex, and frequency. In comparison, the log-normal distribution showed moderate performance, being out-performed by the power logarithmic-normal distribution. Across frequencies, we observed that the goodness of fit becomes more diverse across distribution models at higher frequencies. That is, as the frequency of brain oscillations increases, we could better distinguish models that showed similar performance. Conversely, at lower frequencies, more models demonstrated equal performance.
Conclusions:
We sought to determine which theoretical probability distribution models could best describe the temporal variability of neurophysiological signals. The Generalized Gamma and Exponentiated Weibull distributions consistently outperformed the other 70 models tested, suggesting that they may be the most appropriate models for describing the temporal variability of neurophysiological signals across different populations and types of oscillations. Our analysis also demonstrated that neither power-law scaling alone, nor exponential decay accurately describes the temporal dynamics of brain signals recorded at the level of large neuronal ensembles. Instead, our results suggest that the extreme value theory may offer a more sensitive framework for discriminating between neural states or predicting clinical, behavioral, or cognitive measures from neurophysiological signals.
Lifespan Development:
Aging 2
Modeling and Analysis Methods:
EEG/MEG Modeling and Analysis 1
Methods Development
Keywords:
Aging
Computational Neuroscience
Data analysis
MEG
1|2Indicates the priority used for review
Provide references using author date format
Buzsáki, G. (2014), 'The log-dynamic brain: How skewed distributions affect network operations', Nature Reviews Neuroscience, 15(4), Article 4
Koch, A. L. (1966), 'The logarithm in biology 1. Mechanisms generating the log-normal distribution exactly', Journal of Theoretical Biology, 12(2), 276–290
Mizuseki, K. (2013), 'Preconfigured, skewed distribution of firing rates in the hippocampus and entorhinal cortex', Cell Reports, 4(5), 1010–1021
Roberts, J. A. (2015), 'The heavy tail of the human brain', Current Opinion in Neurobiology, 31, 164–172
Shafi, M. (2007), 'Variability in neuronal activity in primate cortex during working memory tasks', Neuroscience, 146(3), 1082–1108
Shafto, M. (2014), 'The Cambridge Centre for Ageing and Neuroscience (Cam-CAN) study protocol: A cross-sectional, lifespan, multidisciplinary examination of healthy cognitive ageing', BMC Neurology, 14(1), 204
Wang, S. S.-H. (2008), 'Functional Trade-Offs in White Matter Axonal Scaling', Journal of Neuroscience, 28(15), 4047–4056