Physiological noise modeling in fMRI based on the pulsatile component of photoplethysmograph

Abstract

The blood oxygenation level-dependent (BOLD) contrast mechanism allows someone to non-invasively probe changes in deoxyhemoglobin content. As such, it is commonly used in fMRI to study brain activity since levels of 10 deoxyhemoglobin are indirectly related to local neuronal activity through neurovascular coupling. However, the BOLD signal is severely affected by physiological processes as well as motion. Due to this, several noise correction techniques have been developed through the years to correct for the associated confounds. This study sought to refine model-based techniques that utilize the photoplethysmograph (PPG) signal. RETROICOR, a technique commonly used to model fMRI fluctuations induced by cardiac pulsatility was compared with a new technique proposed here, 15 named cardiac pulsatility model (CPM), that is based on convolution filtering. Further, this study investigated whether the variations in the amplitude of the PPG pulses (PPG-Amp) covary with variations in amplitude of pulse-related fMRI fluctuations as well as with systemic low frequency oscillations (SLFOs) present in the global signal (i.e. mean fMRI timeseries averaged across all voxels in gray matter). Capitalizing on 3T fMRI data from the Human Connectome Project, CPM was found to explain significantly more variance in fMRI compared to RETROICOR, 20 particularly for subjects that presented high variance in heart rate during the scan. The amplitude of the fMRI pulserelated fluctuations did not seem to covary with PPG-Amp. That said, PPG-Amp explained significant variance in the GS that did not seem to be attributed to variations in heart rate or breathing patterns. In conclusion, our results suggest that the techniques proposed here can model high-frequency fluctuations due to pulsation as well as lowfrequency physiological fluctuations more accurately than model-based techniques commonly employed in fMRI 25 studies.

Altered Relationship Between Heart Rate Variability and fMRI-Based Functional Connectivity in People With Epilepsy

“…The neural correlates of heart rate found in our work and in Valenza et al ( 29 ) were not entirely consistent, which may result from the more aggressive physiological correction applied in our work. Artifacts due to cardiac pulsatility were removed using the newly proposed cardiac pulsatility model ( 55 ) and systemic low-frequency oscillations were removed through gray matter signal regression ( 47 , 56 ).…”

Physiological and head motion signatures in static and time-varying functional connectivity and their subject discriminability

“…It is well established that head and breathing motion affect areas at the edges of the brain (Jo et al, 2010; Patriat et al, 2015; Satterthwaite et al, 2013), whereas cardiac pulsatility affects areas near the large cerebral arteries just above the neck (Glover et al, 2000; Kassinopoulos and Mitsis, 2020). These observations are based on studies that typically examine the brain regions affected by the aforementioned sources of noise on a voxel-wise basis.…”

A multi-measure approach for assessing the performance of fMRI preprocessing strategies in resting-state functional connectivity

“…Previous studies have shown that the GS derived either by the whole brain or GM are very similar to each other and also that the GS is highly correlated with the mean time series across voxels in WM and CSF (Kassinopoulos & Mitsis, 2019; Power et al, 2017), which further lends support to the idea that WM regressors share common variance with the GS. Furthermore, we observed that the SLFOs that reflect BOLD fluctuations due to changes in heart rate and breathing patterns, and account for a significant fraction of GS fluctuations (Falahpour et al, 2013; Kassinopoulos & Mitsis, 2019, 2020), were well explained using the first 20-30 WM and CSF regressors (Fig. 1).…”

“…Despite the simplicity of GSR, there has been much debate about its use (Liu et al, 2017; Murphy & Fox, 2017). Even though several studies have shown that a large fraction of the GS is associated to physiological processes such as heart rate and breathing activity (Birn et al, 2006; Chang et al, 2009; Falahpour et al, 2013; Kassinopoulos & Mitsis, 2019, 2020; Shmueli et al, 2007; Wise et al, 2004) as well as head motion (Power et al, 2014; Satterthwaite et al, 2013), there is accumulating evidence that GS is also driven by neuronal activity as assessed by intracranial recordings (Schölvinck et al, 2010) and vigilance-related measures (Chang et al, 2016; Falahpour et al, 2018; Liu & Falahpour, 2020; Wong et al, 2013, 2016). Therefore, while our results are in support of GSR for both WM and FIX denoising, we cannot exclude the possibility of removing some neuronal-related fluctuations from the data when the GS is removed.…”

“…Similarly, breathing motion introduces high-frequency artifacts (∼0.3 Hz), mostly at the edges of the brain, that are in phase with the breathing cycles. Slow-frequency fluctuations in heart rate and breathing pattern (<0.1 Hz) are also typically observed during rest which have a direct effect on the cerebral blood flow and levels of oxygenated hemoglobin in the brain (Birn et al, 2006; Chang et al, 2009; Kassinopoulos & Mitsis, 2019, 2020; Shmueli et al, 2007). As a result, they affect widespread regions in the gray matter (GM).…”