Preprocessing Considerations in fMRI Naturalistic Viewing Paradigms

1953 

Abstract Submission 

Menghan Yang¹, Eshin Jolly¹, Luke Chang¹

¹Dartmouth College, Hanover, NH

Menghan Yang  
Dartmouth College
Hanover, NH

Eshin Jolly  
Dartmouth College
Hanover, NH

Luke Chang, PhD  
Dartmouth College
Hanover, NH

Naturalistic stimuli have been increasingly used in brain imaging studies in recent years due to the advantages of enabling researchers to mimic everyday situations, transport individuals into immersive narrative worlds, and evoke neural responses that closely resemble real-life experiences (Chang et al., 2021; Green & Brock, 2002). Compared to resting-state paradigms, naturalistic designs probe a greater variety of cognitive functions (Finn, 2021; Sonkusare et al., 2019), which increases the diversity of mental states (Meer et al., 2020) and the reliability of data (Wang et al., 2017). While naturalistic stimuli have become increasingly popular, most studies rely on traditional preprocessing methods optimized for task or resting-state designs. However, naturalistic viewing paradigms often contain longer range temporal dependencies due to the complexities of the narrative (Hasson et al., 2015) and longer scan times compared to traditional paradigms. If researchers are interested in processes that occur at longer timescales such as emotions (Chang et al., 2021) or memory (Chen et al., 2017), then filtering out slower signals with high pass filters is likely to be undesirable (Lositsky et al., 2016). Therefore, in this study, we examined the impact of different preprocessing steps on intersubject correlations using the Friday Light Nights dataset.

Participants (N=35) watched a 45-minute episode of Friday Night Lights, a character-driven television drama while undergoing fMRI (Chang et al., 2021). We examined the independent effects of different denoising steps on intersubject correlations (ISC) within 50 different regions of interest (ROIs) from a whole brain parcellation (de la Vega et al., 2016). Specifically, we separately examined the impact of despiking, 24 Volterra expanded head motion covariates, 128s high-pass filter, and average CSF activity compared to a baseline detrended data (intercept, linear, and quadratic).

Overall, we observed a similar pattern of temporal ISC results across the brain reported in prior work in our baseline condition (Fig 1A), with higher ISCs exhibited in sensory cortex compared to heteromodal association cortex. Beyond simply detrending the data, despiking and average CSF activity improved temporal ISC for most brain regions, presumably due to decreasing noise variance in the BOLD signal. However, high-pass filtering dramatically decreased ISC, while motion covariates had mixed results across the brain (Fig 1, B, C, and D).
In addition, we also examined spatial ISC within participants by computing pairwise spatial similarity within each ROI across all TRs (Fig 2, A and B). This allowed us to quantify the temporal autocorrelation of spatial patterns for each ROI (Fig 2, C and D). We found that detrending and high-pass filter steps introduced some artifacts in the spatiotemporal signals such as shorter range anti-correlations and longer range correlations, which was particularly notable in the high pass filtering condition. Motion covariates appeared to reduce the impact of these artifacts.

Together these results suggest that different denoising methods exhibit distinct effects on ISC metrics. For ISC, we recommend that researchers include spikes and motion parameters and optionally CSF activity. High pass filtering is likely to have an adverse impact on analyzing naturalistic data and is not recommended. Future work will continue to assess the impact of these preprocessing methods on other measures beyond ISC such as SNR, connectivity based metrics, and scene classification.

Social Neuroscience Other

Activation (eg. BOLD task-fMRI) ²

Motion Correction and Preprocessing ¹

Task-Independent and Resting-State Analysis

FUNCTIONAL MRI

Multivariate

Univariate

Other - Naturalistic stimuli; Preprocessing