Linear Modeling of Neurophysiological Responses to Speech and Other Continuous Stimuli: Methodological Considerations for Applied Research

Abstract

Cognitive neuroscience, in particular research on speech and language, has seen an increase in the use of linear modeling techniques for studying the processing of natural, environmental stimuli. The availability of such computational tools has prompted similar investigations in many clinical domains, facilitating the study of cognitive and sensory deficits under more naturalistic conditions. However, studying clinical (and often highly heterogeneous) cohorts introduces an added layer of complexity to such modeling procedures, potentially leading to instability of such techniques and, as a result, inconsistent findings. Here, we outline some key methodological considerations for applied research, referring to a hypothetical clinical experiment involving speech processing and worked examples of simulated electrophysiological (EEG) data. In particular, we focus on experimental design, data preprocessing, stimulus feature extraction, model design, model training and evaluation, and interpretation of model weights. Throughout the paper, we demonstrate the implementation of each step in MATLAB using the mTRF-Toolbox and discuss how to address issues that could arise in applied research. In doing so, we hope to provide better intuition on these more technical points and provide a resource for applied and clinical researchers investigating sensory and cognitive processing using ecologically rich stimuli.

Neural Measures of Pitch Processing in EEG Responses to Running Speech

“…where X is the time-lagged stimulus feature, λ is a regularization parameter, and Y is the neural response data for a given EEG channel. The regression weights β yield a temporal response function (TRF) that can be interpreted as the stimulus-evoked impulse response from a neural population (Lalor et al, 2009 ; Ding and Simon, 2012 ; Crosse et al, 2021 ). When fitting the regularization parameter in the cortical response analysis, a 3-way nested cross-validation procedure was used in which the data were split into training, validation and test sets as described in Fuglsang et al ( 2020 ).…”

Algorithms for Estimating Time-Locked Neural Response Components in Cortical Processing of Continuous Speech

“…Variations of regularized regression and machine learning methods for estimating TRFs have been previously compared for decoding subject attention in a multi-talker scenario [6], [12], [13]. However, it is unclear how they compare to commonly used sparse TRF estimation techniques such as boosting [14], [15].…”

“…In this work we perform a systematic comparison of TRF algorithms in terms of estimating TRF components. Two widely used TRF estimation algorithms are ridge regression [13], [16] and boosting [3], [14], [15]. The former uses ℓ !…”

“…In this work we perform a systematic comparison of TRF algorithms in terms of estimating TRF components. Two widely used TRF estimation algorithms are ridge regression [13], [16] and boosting [3], [14], [15]. The former uses ℓ 2 regularization which leads to smooth TRFs with broad components, while the latter greedily adds values to the TRF, thereby prioritizing sparsity in the TRF and leading to narrower, sharper components.…”

“…The a-priori parameters may need to be tuned for each predictor type or experiment, or even for each subject Modern TRF analyses involve multiple types of predictors [42] (e.g., envelopes, phoneme onsets, multiple frequency bands for spectrotemporal TRFs). Boosting and banded ridge regression may be suitable for these studies [10], [13], [43], [44]. The component characteristics of TRFs to these higher-level predictors must be determined before SP and EM-SP can be applied.…”