Evaluating RETROICOR and aCompCor-based denoising for improving multi-voxel pattern analysis

Poster Session C - Sunday, April 14, 2024, 5:00 – 7:00 pm EDT, Sheraton Hall ABC

Elizabeth Doss¹, Haley Keglovits¹, Apoorva Bhandari¹, David Badre¹; ¹Cognitive, Linguistic, and Psychological Sciences, Brown University

fMRI measurements from the brain are characterized by small task-driven fluctuations in the background of large amounts of structured noise from different sources. One major source is physiological noise from non-neural sources such as the cardiac and respiratory cycles. RETROICOR is a retrospective correction approach for improving the signal-to-noise ratio of fMRI data employing estimates of the phase of these cycles from concurrent, peripheral physiological measurements. Conversely, aCompCor is a component-based method that utilizes components of measured signals from non-neural anatomical sources such as cerebrospinal fluid for denoising, without the need for physiological measures. However, it remains unclear whether aCompCor or RETROICOR improves the reliability of multivariate pattern analyses, and how these metrics vary across brain regions (Bhandari et al. 2018). Here we evaluate the effects of these denoising pipelines for improving multivariate pattern analyses. We leveraged deep-sampled fMRI and physiology datasets in which participants (N=20) performed a complex decision-making task across 10 fMRI sessions. We examined how including different combinations of denoising methods influence multi-voxel pattern reliability and Representational Dissimilarity Matrix (RDM) reliability. We show that, across the whole brain, neither pattern reliability nor noise ceiling for RDMs are affected by additional RETROICOR denoising if aCompCor is already employed. However, in some selected brain regions, RETROICOR causes a decrease in pattern correlation and noise ceilings. These results suggest that including RETROICOR in a preprocessing pipeline that already includes aCompCor could have a negative effect on the signal to noise ratio in task-based fMRI experiments.

Topic Area: METHODS: Neuroimaging