Shift-invariant multilinear decomposition of neuroimaging data

Abstract

We present an algorithm for multilinear decomposition that allows for arbitrary shifts along one modality. The method is applied to neural activity arranged in the three modalities space, time, and trial. Thus, the algorithm models neural activity as a linear superposition of components with a fixed time course that may vary across either trials or space in its overall intensity or latency. Its utility is demonstrated on simulated data as well as actual EEG, and fMRI data. This work shows how pseudo-multilinear decompositions of multiway data can successfully cope with variable latencies in data derived from neural activitya problem that has caused degenerate solutions especially in modeling neuroimaging data with instantaneous multilinear decompositions. Our algorithm is available for download at www.erpwavelab.org.

Adaptive independent vector analysis for multi-subject complex-valued fMRI data

“…Since different subjects may generate different responses due to variation in response time or in their hemodynamic delay, inter-subject TC variability naturally also exists for multi-subject fMRI data. Kuang et al (2015) proposed a solution by combining shift-invariant CPD (Mørup et al, 2008) and ICA to simultaneously consider inter-subject SM and TC variability. Compared to tensor decomposition, ICAbased analysis extracts subject-specific TCs and/or SMs for emphasizing inter-subject variability.…”

Modeling Sparse Connectivity Between Underlying Brain Sources for EEG/MEG

“…[22,39]). A third field of upcoming research is the extension of our causal model to multi-way data by building on the work of [40]. …”

Modeling latency and shape changes in trial based neuroimaging data

“…Then, the remaining convolutive components should model consistent confounding effects not phase locked to the event. For comparison we included a regular CP analysis as well as the shiftCP model proposed in [16], [24]. The convergence criterion for the algorithm was set as termination when the relative change of the log-posterior was less than 10 −6 or when the algorithm had run for 1000 iterations.…”

“…Each stimulus category included 313 events and an object was presented up to three times. For details on the data set, see also [24]. −500 to 1500 ms forming the data array I = 64 channel × J = 1024 time-points × K = 313 trials.…”

“…The proposed model is a generalization of the model analyzed in [15], [16], [24], which we here will denote shiftCP. The shiftCP model is constrained such that each time profile only has one specific delay value in each trial.…”