Modeling EEG Spectral Features through Warped Functional Mixed Membership Models
Emma Landry, Damla Şentürk, Shafali Jeste, Charlotte DiStefano, Abigail Dickinson, Donatello Telesca · 2024
Read Modeling EEG Spectral Features through Warped Functional Mixed Membership Models by Emma Landry, Damla Şentürk, Shafali Jeste, Charlotte DiStefano, Abigail Dickinson, Donatello Telesca (2024) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2412-08762.txt.
Introduction
Emma Landry and co-authors model children's brainwave spectra with curves that can stretch in time and blend between types, in a 2024 paper. EEG traces from different children misalign: the same feature lands at different frequencies. Older alignment tools assume one shared shape or a few fixed ones. The team's Bayesian model lets each child belong partly to several clusters, with a personal time-warp per curve.
Applied to autism study data, the model cleanly separated the famous one-over-f hum from the alpha rhythm's peak, matching neuroimaging knowledge. A regression layer then priced how age and diagnosis shift the alpha peak's position. Flexible membership, the authors argue, respects real developmental variety.
Catalog record
| CREATOR | Emma Landry, Damla Şentürk, Shafali Jeste, Charlotte DiStefano, Abigail Dickinson, Donatello Telesca |
|---|---|
| YEAR | 2024 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 56 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2412-08762
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Modeling EEG Spectral Features through Warped Functional Mixed Membership Models Emma Landry, Damla Şentürk, Shafali Jeste, Charlotte DiStefano, Abigail Dickinson, Donatello Telesca Abstract A common concern in the field of functional data analysis is the challenge of temporal misalignment, which is typically addressed using curve registration methods. Currently, most of these methods assume the data is governed by a single common shape or a finite mixture of population level shapes. We introduce more flexibility using mixed membership models. Individual observations are assumed to partially belong to different clusters, allowing variation across multiple functional features. We propose a Bayesian hierarchical model to estimate the underlying shapes, as well as the individual time-transformation functions and levels of membership. Motivating this work is data from EEG signals in children with autism spectrum disorder (ASD). Our method agrees with the neuroimaging literature, recovering the 1/f pink noise feature distinctly from the peak in the alpha band. Furthermore, the introduction of a regression component in the estimation of time-transformation functions quantifies the effect of age and clinical designation on the location of the peak alpha frequency (PAF). keywords First keyword , second keyword , and ?? Department of Biostatistics, University of California, Los Angeles, CA 90095 ?? Department of Neurology, Children’s Hospital Los Angeles, Los Angeles, CA 90027 ?? Division of Clinical Psychology, Children’s Hospital Los Angeles, Los Angeles, CA 90027 ?? Center for Autism Research and Treatment, Semel Institute for Neuroscience, University of California, Los Angeles, CA 90095 ?? Corresponding author, ?? 1 Introduction Functional data analysis is a branch of statistics concerned with the analysis of data that comes in the form of random curves. A key characteristic of such data is that it tends to display variation along both the x and y axes, respectively referred to as phase and amplitude variation. This manuscript introduces a Bayesian curve registration method to account for phase variation in normalized data arising from mixed membership models– where individuals partially belong to multiple clusters associated with distinct functional features. Our method is motivated by a case study of resting-state electroencephalography (EEG) measured in a cohort of children with Autism Spectrum Disorder (ASD) and age-matched Typically Developing (TD) children (Fig. 4 ). This data was first introduced, described and analyzed by Dickinson et al. (2018) . In particular, the focus is on spectral analysis in the alpha band (6-12 Hz), within which a noticeable peak occurs. The Peak Alpha Frequency (PAF), defined to be the frequency at which such a peak occurs for an individual, as well as the power associated with the peak, can serve as neuromarkers for ASD, as well as multiple other neurodevelopmental disorders. As a result, the study of the alpha band for resting state EEG has been of notable interest in the field of neuroscience. Deiber et al. (2020) conduct a study on adults with Attention-Deficit Hyperactivity Disorder (ADHD), in which the control group is seen to display higher resting-state alpha power, that is, more prominent peaks. In the case of patients suffering from schizophrenia, Ramsay et al. (2021) note a slower PAF than the controls, i.e. the peak occurs at slower frequencies. Focusing on ASD, Edgar et al. (2015) find a difference in the relationship between age and PAF. Analysis of the data for our case study of interest agrees with those results, noting that TD children display an increase in PAF with age, which their ASD counterparts don’t ( Dickinson et al., 2018 ) . Other authors in the statistical literature have previously tackled the analysis of this data ( Scheffler et al., 2019 ; Shamshoian et al., 2022 ; Marco et al., 2024a ) , but their analyses have not formally considered issues of phase variability to account for mismatch in timing of features across individuals. The registration problem has been extensively explored in the study of functional data. In the frequentist literature, one of the earliest approaches is landmark registration ( Gasser and Kneip, 1995 ) , where certain features of the data are identified to match their timings. Wang and Gasser (1997) ; Wang and Gasser (1999) consider alignment through dynamic time warping, where the shift between two functions is optimized through a cost function. Ramsay and Li (1998) introduce the Procrustes method for alignment, using an iteratively updated cross-sectional mean as a target to estimate time-transformation functions. Many authors have also considered more flexible methods that use self-modelling warping functions, such as through shape-invariant models (SIM) ( Rønn, 2001 ) , or through B-spline basis estimation ( Gervini and Gasser, 2004 ) . More recently, Srivastava et al. (2011) take a geometric approach borrowing from shape analysis methodology and use the Fisher-Rao metric as a distance measure for registration. The work of Telesca and Inoue (2008) introduces the first method that addresses the curve registration problem in the Bayesian framework. They present a hierarchical model to formally account for amplitude and phase variation. For some other recent development in Bayesian curve registration, see Claeskens et al. (2010) , Cheng et al. (2016) , Bharath and Kurtek (2020) , Matuk et al. (2022) . While our contribution is novel in that we consider the framework of mixed membership models, multiple authors have allowed for curve registration methods to incorporate the concept of groups, or features. In particular, several strategies for clustering of functional data while addressing the alignment problem have been developed ( Liu and Yang, 2009 ; Zhang and Telesca, 2014 ; Tang et al., 2023 ) . Other approaches consider the case where data may vary in multiple functional directions, such as Kneip and Ramsay (2008) who address it through functional principal components, or Earls and Hooker (2017) that combine curve registration with factor analysis. In this paper, we extend Bayesian hierarchical curve registration introduced by Telesca and Inoue (2008) to mixed membership models ( Erosheva et al., 2004 ) .This class of models assumes the existence of a finite number of subpopulations, to which each individual belongs to varying degrees. Marco et al. (2024a) introduce mixed membership to the field of functional data analysis, where the authors exploit the multivariate Karhunen-Loève theorem to extend finite Gaussian process (GP) mixture models. Our proposed model addresses the issue of phase variability, which arises from the fact that individual functions may operate on a distorted timescale, leading to misalignment of functional data in physical time ( Ramsay and Silverman, 1997 ) . We leverage the work of Brumback and Lindstrom (2004) to model the time-transformation functions in an unconstrained fashion and transform them so that shape restrictions are satisfied. This allows for flexible modeling, including covariate adjustments, and the estimation through B-splines is conveniently incorporated into MCMC sampling. The remainder of the paper is organized as follows. Section 2 introduces the curve registration model for the mixed membership framework– including discussion of prior choices, identifiablity concerns, and posterior inference details. Section 3 illustrates our method in practice. In section 3.1 we present a simulation study to assess model fit, and explore the effects of semi-supervision through partial labelling of observations. Section 3.2 applies the model to electroencephalography (EEG) data. Finally, section 4 provides a critical discussion of our methodology. 2 Methods 2.1 Joint registration and mixed membership