Modeling latency and shape changes in trial based neuroimaging data

Abstract

Abstract-To overcome poor signal-to-noise ratios in neuroimaging, data sets are often acquired over repeated trials that form a three-way array of space×time×trials. As neuroimaging data contain multiple inter-mixed signal components blind signal separation and decomposition methods are frequently invoked for exploratory analysis and as a preprocessing step for signal detection. Most previous component analyses have avoided working directly with the tri-linear structure, but resorted to bi-linear models such as ICA, PCA, and NMF. Multi-linear decomposition can exploit consistency over trials and contrary to bi-linear decomposition render unique representations without additional constraints. However, they can degenerate if data does not comply with the given multi-linear structure, e.g., due to time-delays. Here we extend multi-linear decomposition to account for general temporal modeling within a convolutional representation. We demonstrate how this alleviates degeneracy and helps to extract physiologically plausible components. The resulting convolutive multi-linear decomposition can model realistic trial variability as demonstrated in EEG and fMRI data.

Multi-subject fMRI analysis via combined independent component analysis and shift-invariant canonical polyadic decomposition

“…For example, the subject-specific intensities have been utilized as predictors of disease risk in numerous studies across a wide-variety of illnesses (Caffo, et al, 2010;Li, et al, 2012;Calhoun et al, 2004;Calhoun et al, 2009;Tursich et al, 2015;Weiland et al, 2015). Note we did not consider TC shape variability (Mørup et al, 2011) across multiple subjects, future work on this extension might provide additional and valuable inter-subject variability. Tables Table 1. The t-values and p-values of the paired t-test for evaluating the difference between the average 1 c ρ − values of the proposed approach and tensor PICA/SCP_O/SCP_I, for the influence of SM and TC changes and noise on the task-related SM and TC estimates, as displayed in Fig.…”

“…As brain function is still poorly understood, and fMRI data are noisy, datadriven approaches exhibit great potential in extracting spatial and temporal components from fMRI data with little to no prior information about the brain. Among others, independent component analysis (ICA) (McKeown et al, 1998;Vigario and Oja, 2008;Calhoun et al, 2001;Guo and Pagnonib, 2008;Du and Fan, 2013;Lee et al, 2008;Michael et al, 2014;Risk et al, 2014) and tensor decomposition (TD) (Andersen and Rayens, 2004;Beckmann and Smith, 2005;Mørup et al, 2008;Cichocki et al, 2009;Cichocki et al, 2014;Mørup et al, 2011), two key approaches of blind source separation (BSS), have provided promising results in multi-subject fMRI analysis.…”

“…Among others, canonical polyadic decomposition (CP decomposition, or CPD), also called CANDECOMP, or parallel factor analysis (PARAFAC), decomposes multiway data into a linear combination of a series of outer product of rank-one tensors. CPD is a popular TD method as it is not only able to provide a unique decomposition under mild conditions, but also provides useful results for fMRI data (Andersen and Rayens, 2004;Beckmann and Smith, 2005;Stegeman, 2007;Mørup et al, 2008;Mørup et al, 2011;Helwig and Hong, 2012). When applied to multi-subject fMRI analysis, CPD is able to obtain shared SMs, shared TCs, and subject-specific intensities (Beckmann and Smith, 2005;Stegeman, 2007;Kuang et al, 2013).…”

A tutorial on the use of temporal exploratory factor analysis in developmental ERP research – chances and challenges

“…The trilinear model has been proposed as an advancement of the EFA model which acknowledges the role of electrode sites and sampling points as measurement occasions instead of treating one or the other as "observations" (Achim & Bouchard, 1997;Cong et al, 2015;Möcks, 1988bMöcks, , 1988aWang et al, 2000). There have even been proposals how latency shifts can be directly acknowledged within such a model (Harshman et al, 2003;Harshman & Lundy, 1994;Mørup et al, 2008Mørup et al, , 2011. Apart from modifications of the EFA model to ERP data, it has been suggested that EFA can also be helpful when conducted on typical transformations of ERP data such as scalp current density (Kayser & Tenke, 2015b, 2015a, time-frequency decompositions (Barry et al, 2019), or wavelet-transformations (Mørup et al, 2006b).…”

Tracing Network Evolution Using The Parafac2 Model

“…Previously, for the analysis of time-evolving data, a popular ten-sor factorization model called the CANDECOMP/PARAFAC (CP) [5,6] model has been used to extract temporal patterns to address the temporal link prediction problem [7], to capture the evolving popularity of different meaningful topics from email threads [8] and to detect suspicious activity in network traffic [9]. However, the CP model assumes that underlying patterns stay the same across time, which may not be satisfied by dynamic data [10,11,12,13].…”