Synopsis
Through a series of complex processes, under the umbrella term of neurovascular coupling, neuronal activity ultimately manifests as a signal change in an MR image via the blood-oxygenation level dependent (BOLD) contrast. Functional MRI (fMRI) capitalises on this contrast mechanism to infer neuronal activity from BOLD contrast variation in a time series, typically acquired while the participant engages in a task. This approach has proved valuable in furthering our understanding of the working of the human brain. Here, issues pertinent to acquiring data with sufficiently high sensitivity to detect such changes are considered, e.g. susceptibility effects, physiological noise and approaches facilitating high spatio-temporal resolution.
Typical Data Acquisition
The most commonly used acquisition scheme for fMRI is 2D gradient echo echo-planar imaging (EPI). This approach offers high sensitivity and acquires an image in a very short period of time (e.g. 50-100ms). Slices are acquired sequentially with the time taken to acquire a single image volume dependent on the spatial coverage that is required, e.g. in the region of 3s for whole brain coverage with 3mm isotropic resolution. The modest spatio-temporal resolution achieved in this scenario is largely sufficient, given the temporally smooth characteristics of the canonical haemodynamic response function (HRF).
Consequences of Low Bandwidth
The excellent speed of EPI-based approaches has made it a popular choice but comes at the price of a relatively low bandwidth along the phase-encoded direction. This makes the approach sensitive to any sources of field inhomogeneity, which leads to differential precessional frequency. For this reason, fat saturation techniques are required. In addition, any intrinsic susceptibility differences lead to field inhomogeneities, which result in signal distortions, dropouts, or both. The impact of susceptibility-related distortions will depend on the duration of the EPI readout, with longer readouts, of lower bandwidth, suffering from greater distortions. Intra-voxel dephasing resulting from these susceptibility gradients also causes signal loss, with the degree of signal dropout in the image depending on the echo time chosen. It has been shown that the optimal echo time at which BOLD sensitivity is maximised is when it matches the T2* of the tissue of interest (Deichmann et al. 2002).Optimisation to Mitigate Susceptibility Effects
A number of approaches can be adopted to mitigate susceptibility-related distortions. Among these are the use of parallel imaging to reduce the duration of the readout; angulation of the slices to distribute, and therefore reduce, the impact of the susceptibility gradients; or the application of additional z-shim gradients that counteract the inherent susceptibility gradients (e.g. Deichmann et al. 2002, Weiskopf et al. 2006). As is typically the case for MRI, optimisation for one region comes with a cost for another. For example, choosing an optimal z-shim value for one region will cause dephasing in unaffected areas. Therefore the spatial areas of interest should be considered carefully when selecting the most appropriate sequence setup. An alternative approach gaining popularity is to acquire multi-echo data, with an array of echo times, which can subsequently be combined to improve overall BOLD sensitivity (Poser et al. 2006).
Advanced Acquisitions for High Spatio-temporal Resolution
More recent advances in acquisition techniques, e.g. multiband (or simultaneous multi-slice) imaging (Xu et al. 2013, Setsompop et al. 2012) or 3D-EPI with CAIPIRINHA sampling (Narsude et al. 2016), have enabled significant increases in spatio-temporal resolution. Increased temporal resolution is of benefit for real-time applications such as neurofeedback or for improved sampling (and subsequent removal) of physiological noise. Increased spatial resolution benefits applications requiring enhanced spatial specificity, e.g. to examine hippocampal sub-fields (Zeidman et al. 2015) or to again mitigate the effects of physiological noise (which scales down with voxel size, Triantafyllou et al. 2005). It should be noted that such acceleration approaches, which exploit coil information in order to unfold aliased voxels, will have greater sensitivity to motion throughout the acquisition and in particular motion during the acquisition of the required calibration data.
Use of Ultra-high Field
Another means by which spatial specificity is gained is by capitalising on the enhanced sensitivity afforded by ultra-high field (≥7T). This raises the possibility of identifying discrete units of neuronal computation such as layers, columns or stripes (van der Zwaag et al. 2015, Koopmans et al. 2011). For example, colour and disparity-selective columns have been identified in the visual cortex (Nasr et al. 2016). The use of ultra-high field has been combined with spin-echo approaches to increase sensitivity to micro-vasculature with a view to further enhancing spatial specificity (de Martino et al. 2013). Again investigating the visual cortex, this approach has enabled the mapping of orientation-preference columns (Yacoub et al. 2008).
General Acquisition Considerations
Motion must always be carefully considered and mitigated against in functional studies. Retrospective correction of bulk motion, via rigid-body co-registration, can help to realign data but cannot correct for effects such as position-specific receive field modulation or field inhomogeneities. Therefore careful control to minimise any potential motion, e.g. via participant coaching and motion restriction are vital. In addition, physiological effects of breathing and pulsation can induce signal fluctuations thereby degrading the functional sensitivity of the time series. A number of approaches have been developed to correct these effects, e.g. RETROICOR (Glover et al. 2000) and DRIFTER (Sarkka et al. 2012), and increase functional sensitivity (e.g. Jorge et al. 2013, Lutti et al. 2013).Acknowledgements
The Wellcome Trust Center for Neuroimaging is supported by core funding from the Wellcome Trust Grant no. 0915/Z/10/Z.References
Deichmann et al. NeuroImage 2002, 15(1): 120-135
Weiskopf et al. NeuroImage 2006, 33(2): 493-504
Poser et al. MRM 2006, 55(6): 1227-1235
Setsompop et al., Magn Reson Med 2012, 67(5):1210-1224
Xu et al. NeuroImage 2013, 83:991-1001
Narsude et al., Magn Reson Med 2016, 75(6): 2350-2361
Zeidman et al., Cortex 2015, 73:240-256
Triantafyllou et al., NeuroImage 2005, 26(1): 243-250
De Martino et al. PLoS One 2013, 8(3): 30-32
Yacoub et al. PNAS 2008, 105(30): 10607-10612
van der Zwaag et al. NMR in Biomed 2015, 29: 1274-1288
Koopmans et al. NeuroImage 2011, 56(3): 1276-1285
Nasr et al. J. Neurosci. 2016, 36(6): 1841-1857
Glover et al. Magn Reson Med 2000, 44(1): 162-167
Sarkka et al.NeuroImage 2012, 60:1517-1527.
Jorge et al. Magn Reson Med 2013, 31(2): 212-220
Lutti et al. Magn Reson Med 2013, 69: 1657-1664