Impact of hemodynamic effects on diffusion-weighted fMRI signals

Abstract

In some recent studies, diffusion weighted functional MRI has been proposed to provide contrast immune to vascular changes. Increases in relative signal change during neuronal activation observed under increasing diffusion weighting support the possible diffusion based origin of this contrast. A recent diffusion tensor imaging (DTI) study has also reported the use of Fractional Anisotropy (FA) to track activation in white matter. In this study we aimed to establish if relatively high diffusion weighting (b = 1200 and 1800 s/mm²) eliminates the strong vascular influences brought about by 100% O₂ and carbogen (95%O₂ + 5% CO₂) induced vascular challenges in gray matter (GM) and white matter (WM) of rat brain. We also aimed to characterize the influences of these vascular changes on FA, both in GM and in WM. Our study endorses previous reports that even relatively heavily diffusion weighted data can be significantly influenced by hemodynamic changes. However, this was not only observed in GM, but also in WM. Moreover, our study demonstrates that the estimator used to calculate the relative changes should be carefully chosen in order to avoid biases at low signal-to-noise ratios (SNRs) which accompany increasing diffusion weighting. With the use of robust estimators, we found no increases in relative change with increasing b-value during both vascular challenges. Our data also demonstrate that FA can be significantly influenced by hemodynamics, both in GM and in WM. The observed influence of diffusion weighting direction on relative signal change in GM was shown to be associated with structural differences among various regions. If diffusion based functional contrasts immune to hemodynamics do exist, our results highlight the difficulty in discerning those diffusion changes from accompanying vascular changes.

Introduction

Functional MRI (fMRI) has revolutionized neuroimaging by providing unprecedented opportunities for non-invasive assessment of brain function. This has empowered a growing list of neuroscience studies, which hitherto seemed implausible. While the high sensitivity of fMRI techniques based on blood oxygenation level-dependent (BOLD), cerebral blood flow (CBF) or cerebral blood volume (CBV) signals has been a key to this success, the fact that fMRI relies on metabolically driven hemodynamic responses which are only indirectly related to neuronal activity (Logothetis, 2003), generally results in limited spatial and temporal specificity. Though there have been reports of achievable sub-millimeter scale spatial specificity (Duong et al., 2001, Kim et al., 2000) and means to work-around the low temporal resolution of these signals (Menon and Kim, 1999), development of methodologies to improve on these aspects is an ongoing research topic. One of the recurring themes in this regard has been the use of diffusion weighted BOLD fMRI (DW fMRI) scans.

To begin with, fMRI studies with diffusion weighted scans were mostly performed to assess the composition of BOLD signals and the implications of changing various factors influencing it (Boxerman et al., 1995, Duong et al., 2003, Lee et al., 1999, Song et al., 1996, Zhong et al., 1998). It was observed that at 1.5 T a significant portion of BOLD signal arises from protons within the vascular space or cerebro-spinal fluid (CSF) (Song et al., 1996). However, what is of relevance to studies on brain function is mostly the contributions from capillary beds, as these co-localize best with the site of neuronal activity (Menon et al., 1995).

DW fMRI techniques could be used to suppress most of the intravascular contributions from large vascular structures and thus improve the spatial specificity, in particular, with spin echo acquisitions at high field strengths (Duong et al., 2003, Lee et al., 2002). Furthermore, the observation that apparent diffusion coefficients (ADC) of the extravascular tissue are dependent on the intravascular susceptibility (Does et al., 1999, Zhong et al., 1991) hinted at the possibility of using ADC changes to measure hemodynamic activation responses. Because of the sensitivity of ADC to flow changes, it has been argued that ADC would be sensitive to small arterial networks and capillaries, since turbulent flows mostly occur in the arteries and not in the veins (Song et al., 2002). With this idea, the use of ADC changes was suggested as complementary to BOLD changes (Song and Li, 2003, Song et al., 2002, Song et al., 2003). The authors showed that overlapping BOLD and ADC activation regions would correspond to capillary and venule components and thus would be more specific to neuronal activation. Besides, the observation that ADC changes preceded BOLD signal changes by about 1 s (Gangstead and Song, 2002, Harshbarger and Song, 2004) gave further credence to the idea that the combined contrast improved the detection of stimulus induced changes around small arterial and capillary networks.

While the above mentioned studies on ADC changes were performed at low diffusion weighting (b-value < 205 s/mm²), another study performed at higher b-values (≈ 200, 1400 s/mm²) reported functional ADC decreases that could have resulted from transient cortical cell swelling and not from vascular effects (Darquie et al., 2001). This would be more directly associated to neuronal activation and was a marked departure from the widely held views about possible sources for signal changes, opening up new vistas in search of contrasts not based on hemodynamic responses. In other studies, novel temporal signatures were observed using DW fMRI, further strengthening the possible non-hemodynamic origins of those changes (Le Bihan, 2007, Le Bihan et al., 2006). A recent diffusion tensor imaging (DTI) study has reported the possibility of using Fractional Anisotropy (FA) changes for the detection of neuronal activation in white matter (Mandl et al., 2008). This fDTI study not only emphasized the possible existence of a contrast based purely on cellular changes, but also extended functional studies to white matter, which has been rarely reported in BOLD fMRI studies.

While understanding the origins of the proposed diffusion based contrasts is still a work in progress, an issue that has persistently dogged these contrasts is the question as to how immune these are to accompanying vascular effects. Experiments showed that in studies focusing on gray matter, it would be hard to separate vascular and cellular changes, even at high b-values (Goerke and Moller, 2007, Jin and Kim, 2008, Miller et al., 2007, Yacoub et al., 2008), with most studies asserting that the likely cause of ADC changes is vascular in nature. However, there are still some observations from these studies which cannot be easily explained by assigning vascular origins to these changes. e.g., the observed increasing percent signal change in diffusion weighted scans with increasing b-values (Le Bihan et al., 2006, Miller et al., 2007), or the differing temporal characteristics of the observed ADC (or mean diffusivity (MD)) changes (Le Bihan et al., 2006). Also, there is little known as to how diffusion based parameters like FA would work as functional contrasts or about their possible extensions to studies on white matter activation.

This study revisits some of these issues and tries to elucidate the possible interactions between vascular changes and DW fMRI and fDTI signals, both in gray and white matter. Specifically, we investigated a) If voxel intensities in gray and white matter are affected by vascular (CBF and or CBV) changes during DW fMRI scans, and b) To what extent diffusion parameters MD and FA are affected by vascular changes. Besides these, we also address possible issues in estimating the changes that may be present. Therefore, an MRI experiment using relatively high b-values and six non-collinear DW directions was performed in rats undergoing respiratory transitions from air to oxygen to carbogen (5%CO₂ + 95%O₂), which induces strong vascular responses (Lu et al., 2009, Moonen and Bandettini, 2000).