Elsevier

NeuroImage

Volume 168, March 2018, Pages 366-382
NeuroImage

The impact of ultra-high field MRI on cognitive and computational neuroimaging

https://doi.org/10.1016/j.neuroimage.2017.03.060Get rights and content

Highlights

  • Increasing field strength provides advantages in studying brain function in vivo.

  • High fields allow imaging functional subdivisions of sub-cortical regions.

  • At high fields functional imaging can be performed at the mesoscopic scale.

  • At the mesoscale brain inspired computational models can be tested and developed.

  • Laminar neourvascular coupling models need updating to extend applications.

Abstract

The ability to measure functional brain responses non-invasively with ultra high field MRI (7 T and above) represents a unique opportunity in advancing our understanding of the human brain. Compared to lower fields (3 T and below), ultra high field MRI has an increased sensitivity, which can be used to acquire functional images with greater spatial resolution, and greater specificity of the blood oxygen level dependent (BOLD) signal to the underlying neuronal responses.

Together, increased resolution and specificity enable investigating brain functions at a submillimeter scale, which so far could only be done with invasive techniques. At this mesoscopic spatial scale, perception, cognition and behavior can be probed at the level of fundamental units of neural computations, such as cortical columns, cortical layers, and subcortical nuclei. This represents a unique and distinctive advantage that differentiates ultra high from lower field imaging and that can foster a tighter link between fMRI and computational modeling of neural networks.

So far, functional brain mapping at submillimeter scale has focused on the processing of sensory information and on well-known systems for which extensive information is available from invasive recordings in animals. It remains an open challenge to extend this methodology to uniquely human functions and, more generally, to systems for which animal models may be problematic. To succeed, the possibility to acquire high-resolution functional data with large spatial coverage, the availability of computational models of neural processing as well as accurate biophysical modeling of neurovascular coupling at mesoscopic scale all appear necessary.

Introduction

Since its introduction (Bandettini et al., 1992, Kwong et al., 1992, Ogawa et al., 1992), functional magnetic resonance imaging (fMRI) has taken the field of cognitive neuroscience by storm. Here we aim at highlighting the potential for imaging brain function at ultra high fields (UHF, 7 T and above). In more than twenty years of fMRI research, our understanding of the complex physiological changes that underlie the fMRI signal (i.e. the neurovascular coupling and the hemodynamic response (Uludag et al., 2009)) has considerably advanced. As a result, the confidence in interpreting the fMRI signal has increased and fMRI has become the method of choice for many studies of basic sensory processing (e.g. visual, auditory and somatosensory) and higher cognitive functions (e.g. memory, language, emotions) (see Fig. 1). Other factors contributing to the ever increasing number of fMRI applications have been the ability to map human brain function at a high spatial resolution non-invasively with large spatial coverage and the availability of research dedicated MRI scanners. Using fMRI (at a spatial resolution of 8 mm3 or lower) it has been possible to investigate preferential responses of cortical areas for specific stimulus categories (Belin et al., 2000, Kanwisher et al., 1997) and the coarse large-scale topographic representation of sensory features (Sereno et al., 1995).

As a result of continuous technical advances (Moeller et al., 2010, Setsompop et al., 2012, Xu et al., 2013), functional imaging at conventional field strengths (≤3 T) can now be performed at a resolution of 8 mm3 (a threefold improvement compared to early fMRI studies) with sub-second temporal resolution (see e.g. protocols at http://www.humanconnectome.org/) (Ugurbil et al., 2013). In spite of these advances, the spatial resolution of fMRI at conventional field strengths still poses limitations on the study of the human brain. Inspired by histological studies (y Cajal, 1995), invasive electrophysiology has shown that the cortex is organized in distinct cortical layers with unique functional properties (e.g. feed-forward processing and feedback processing) and that cortical areas are composed of processing units that share preference for one stimulus feature across the cortical depth (i.e. cortical columns) (Hubel and Wiesel, 1968, Mountcastle, 1997). The role of subcortical structures and their subdivisions in both feed-forward and feedback processing has also been extensively investigated (e.g. the role of the inferior colliculus in processing sound information; (FitzPatrick, 1975, Merzenich and Reid, 1974, Winer and Schreiner, 2005). The plethora of information coming from invasive research suggest that the mesoscopic organizational level of the brain represents an intermediate comprehensive level of analysis, at which information from single neurons (microscopic scale) and large-scale areal function (macroscopic scale) meet (Mitra, 2014) (Fig. 2).

The continuous development of fMRI techniques has been paralleled by the development of analysis strategies. Functional responses have been analyzed to link behavioral, cognitive and perceptual changes to the measured fMRI signals by investigating differences between experimental conditions in single voxels (i.e. the General Linear Model [GLM]) (Friston et al., 1995) or by considering the information in spatial patterns of responses (i.e. multivoxel pattern analysis [MVPA], fMRI-decoding, representational similarity analysis [RSA]) (Cox and Savoy, 2003, De Martino et al., 2008, Formisano et al., 2008, Haxby et al., 2001, Kriegeskorte et al., 2008, Pereira et al., 2009, Staeren et al., 2009) and in functionally connected brain networks (Greicius et al., 2003). More recently, several analysis techniques (population receptive field [pRF] mapping, fMRI encoding and model based decoding) (Dumoulin and Wandell, 2008, Kay et al., 2008, Naselaris et al., 2010, Wandell, 1999, Wandell and Winawer, 2015) have been introduced in order to link fMRI signals with computational models of brain function.

Analysis techniques such as MVPA or RSA have been proposed to leverage the mesoscopic scale information distributed over multiple large (>1 mm isotropic) voxels (Haynes and Rees, 2005, Kamitani and Tong, 2005). However, the source of this information remains uncertain (see section on “Analyzing patterns of fMRI responses: MVPA and RSA” for more details). At conventional field strengths, only a few studies (Koopmans et al., 2010, Ress et al., 2007) have attempted mapping (i.e. spatially localize) non-invasively the human brain at the mesoscopic scale (Fig. 2).

In other words, linking computations of single columns and layers to imaging data collected in vivo is a task that necessitates high spatio-temporal resolution with sufficient temporal signal to noise ratio (tSNR), which is typically not available at conventional field strengths. Similar limitations apply when imaging functional properties within small subcortical structures and the cerebellum. For technical reasons (e.g. the properties and geometry of receive coils), the sensitivity of fMRI is higher in the cortex of the human brain compared to the thalamus or the brainstem. This poses additional constraints when imaging subcortical structures for which high resolution is essential given their small size.

Increasing field strength is accompanied by increases in the image signal to noise ratio (SNR) (Pohmann et al., 2016; Ugurbil et al., 2003; Vaughan et al., 2001) and in the BOLD contrast. In turn these can be traded for increasing the spatial resolution, up to 0.7 mm3 (e.g. 0.9 mm isotropic) in a large field of view (Vu et al., 2016a) or even 0.27 mm3 in a small visual region (Heidemann et al., 2012). Together with this increase in functional SNR, modeling studies have shown that higher magnetic fields are additionally advantageous as the BOLD signal becomes more sensitive to the microvasculature (Uludag et al., 2009) and in turn more accurate with respect to the site of neuronal activity. The increased SNR does not necessarily translate to an increased tSNR, the relation between these two is dependent on the impact of physiological and thermal noise (Triantafyllou et al., 2005). In low SNR regimes (e.g. as when acquiring high spatial resolution functional images), thermal noise dominates and increasing SNR results in increased tSNR. In high SNR regimes dominates (e.g. moderate to low spatial resolution acquisitions), physiological noise dominates and correction strategies to limit the impact of this source of noise have to be considered (Hutton et al., 2011).

The high SNR available at UHF enables alternative imaging strategies (i.e. non BOLD based measures) such as cerebral blood volume (CBV) based fMRI (Lu et al., 2003). It has been proposed that CBV-based fMRI is more specific to the underlying neuronal activation in comparison to gradient echo (GE)- based BOLD fMRI (Zhao et al., 2005, Zhao et al., 2006) and initial promising results for non invasive measurements in humans have been reported (Guidi et al., 2016, Huber et al., 2015, Huber et al., 2016) (for a review see (Lu and van Zijl, 2012)).

In the last few years, recognition of the advantages of UHF has promoted a gradual shift to imaging the brain at field strengths higher than 3 T (see Fig. 1 and (van der Zwaag et al., 2016)). However, despite the efforts in several research groups, functional imaging at UHF remains a niche. Of course, the availability of UHF MR scanners is a major limiting factor. But normalizing the number of publications by the number of available scanners does not account completely for the differences observed between the numbers of studies conducted at 3 T and at 7 T (see Fig. 1). Thus, the open technical challenges of UHF and, especially, the availability of instruments that provide easy-to-use solutions to these technical challenges appear to be still factors limiting the transition to higher field (≥7 T) scanners for human cognitive neuroscience investigations. As UHF scanner availability is increasing and new developments are steadily mitigating the open technical challenges, the field of cognitive neuroscience is likely ready for a definitive paradigm shift.

Many reviews have presented discussed the advantages and disadvantages of different imaging strategies at UHF (see e.g. (Turner, 2016; Ugurbil, 2016; van der Zwaag et al., 2016)). Here we tie methodological considerations on analysis techniques such as univariate statistical modeling, MVPA, RSA and computational modeling to UHF imaging. We present concrete examples that we draw from the study of the (computational) processes underlying the perception of sounds (and its modulation by cognitive demands). Finally, we offer an overview on the current state of functional imaging at UHF by reviewing applications to higher cognitive functions in the study of both cortical and subcortical processing and try to outline the current limits in the interpretability of UHF fMRI data.

Section snippets

The nature of BOLD fMRI

Since its introduction, a main criticism of fMRI has been its indirect link to neuronal activity (Logothetis, 2008, Ugurbil, 2016, Uğurbil et al., 2003). Signals measured by fMRI reflect vascular and metabolic consequences of neuronal activity and their accuracy with respect to the neuronal processes of interest (i.e. specificity) is strongly dependent on the way that these vascular/metabolic changes are measured. As such, the spatial specificity of fMRI signals depends on both physiology and

Analyzing functional responses

Since its introduction, fMRI provided neuroscientists with an unprecedented amount of data. Even at a resolution of 3 mm a functional scan of the whole brain contains tens of thousands of cortical voxels (i.e. gray matter voxels). For each voxel, responses can be collected with a temporal resolution that nowadays reaches the hundreds of milliseconds (Ugurbil et al., 2013). For this reason the development of fMRI acquisition techniques has been paralleled by a continuous and steady development

fMRI at ultra high fields, example applications

In what follows we will present an overview of 7 T fMRI applications of both cortical and subcortical processing. Examples of cortical applications are further subdivided according to imaging resolution and cognitive functions including basic perception. A more specific focus on perception can be gathered from the review of Dumoulin et al. (submitted to this special issue).

Discussion and future outlook

Functional magnetic resonance imaging has revolutionized the study of human brain function. After more than twenty years of investigations, the shear number of applications of fMRI to the study of perception and cognition is a testament to the utility of the method (Fig. 1). The main criticism to fMRI, its indirect link to neuronal activity, is slowly being resolved through simultaneous recordings of fMRI and electrophysiology (Logothetis et al., 2001) and invasive animal (optical) recordings (

Acknowledgements

This work was supported by the Netherlands Organization for Scientific Research (NWO; VIDI grant 864-13-012 to F.D.M., VENI grant 451-15-012 to M.M. and VICI grant 453-12-002 to E.F).

References (198)

  • S.O. Dumoulin et al.

    Population receptive field estimates in human visual cortex

    Neuroimage

    (2008)
  • T.C. Emmerling et al.

    Decoding the direction of imagined visual motion using 7T ultra-high field fMRI

    Neuroimage

    (2016)
  • O.K. Faull et al.

    Functional subdivision of the human periaqueductal grey in respiratory control using 7 T fMRI

    Neuroimage

    (2015)
  • E. Formisano et al.

    Mirror-symmetric tonotopic maps in human primary auditory cortex

    Neuron

    (2003)
  • A. Fracasso et al.

    Systematic variation of population receptive field properties across cortical depth in human visual cortex

    Neuroimage

    (2016)
  • A. Gardumi et al.

    The effect of spatial resolution on decoding accuracy in fMRI multivariate pattern analysis

    Neuroimage

    (2016)
  • J. Goense et al.

    High-resolution fMRI reveals laminar differences in neurovascular coupling between positive and negative BOLD responses

    Neuron

    (2012)
  • J.B.M. Goense et al.

    Laminar specificity in monkey V1 using high-resolution SE-fMRI

    Magn. Reson. Imaging

    (2006)
  • M. Guidi et al.

    Lamina-dependent calibrated BOLD response in human primary motor cortex

    Neuroimage

    (2016)
  • K.V. Haak et al.

    Connective field modeling

    Neuroimage

    (2013)
  • M. Havlicek et al.

    Physiologically informed dynamic causal modeling of fMRI data

    Neuroimage

    (2015)
  • J. Heinzle et al.

    A hemodynamic model for layered BOLD signals

    Neuroimage

    (2016)
  • M.B. Hoffmann et al.

    Retinotopic mapping of the human visual cortex at a magnetic field strength of 7T. Clinical neurophysiology

    Off. J. Int. Fed. Clin. Neurophysiol.

    (2009)
  • L. Huber et al.

    Cortical lamina-dependent blood volume changes in human brain at 7 T

    Neuroimage

    (2015)
  • L. Huber et al.

    Functional cerebral blood volume mapping with simultaneous multi-slice acquisition

    Neuroimage

    (2016)
  • Y. Kajikawa et al.

    How local is the local field potential?

    Neuron

    (2011)
  • K.N. Kay et al.

    Attention reduces spatial uncertainty in human ventral temporal cortex

    Curr. Biol.

    (2015)
  • B.P. Klein et al.

    Attraction of position preference by spatial attention throughout human visual cortex

    Neuron

    (2014)
  • P. Kok et al.

    Selective Activation of the Deep Layers of the Human Primary Visual Cortex by Top-Down Feedback

    Curr. Biol.

    (2016)
  • P.J. Koopmans et al.

    Multi-echo fMRI of the cortical laminae in humans at 7T

    Neuroimage

    (2011)
  • P.J. Koopmans et al.

    Whole brain, high resolution spin-echo resting state fMRI using PINS multiplexing at 7 T

    Neuroimage

    (2012)
  • N. Kriegeskorte et al.

    How does an fMRI voxel sample the neuronal activity pattern: compact-kernel or complex spatiotemporal filter?

    Neuroimage

    (2010)
  • H. Lu et al.

    A review of the development of Vascular-Space-Occupancy (VASO) fMRI

    Neuroimage

    (2012)
  • A. Alink et al.

    fMRI orientation decoding in V1 does not require global maps or globally coherent orientation stimuli

    Front Psychol.

    (2013)
  • P. Andersson et al.

    Real-time decoding of brain responses to visuospatial attention using 7T fMRI

    PLoS One

    (2011)
  • P.A. Bandettini et al.

    Time course EPI of human brain function during task activation. Magnetic resonance in medicine

    Off. J. Soc. Magn. Reson. Med./Soc. Magn. Reson. Med.

    (1992)
  • M. Barth et al.

    Very high-resolution three-dimensional functional MRI of the human visual cortex with elimination of large venous vessels

    NMR Biomed.

    (2007)
  • M.A. Batson et al.

    Single session imaging of cerebellum at 7 T: obtaining structure and function of multiple motor subsystems in individual subjects

    PLoS One

    (2015)
  • P. Belin et al.

    Voice-selective areas in human auditory cortex

    Nature

    (2000)
  • D. Berron et al.

    Strong Evidence for Pattern Separation in Human Dentate Gyrus

    J. Neurosci.

    (2016)
  • M. Bianciardi et al.

    In vivo functional connectome of human brainstem nuclei of the ascending arousal, autonomic, and motor systems by high spatial resolution 7-Tesla fMRI

    Magma

    (2016)
  • S. Bode et al.

    Tracking the unconscious generation of free decisions using ultra-high field fMRI

    PLoS One

    (2011)
  • J.L. Boxerman et al.

    The intravascular contribution to fMRI signal change: monte Carlo modeling and diffusion-weighted studies in vivo

    Magn. Reson. Med.

    (1995)
  • J. Budde et al.

    Functional MRI in human subjects with gradient-echo and spin-echo EPI at 9.4 T

    Magn. Reson. Med.

    (2014)
  • S. Da Costa et al.

    Tuning in to sound: frequency-selective attentional filter in human primary auditory cortex

    J. Neurosci.

    (2013)
  • F. De Martino et al.

    Whole brain high-resolution functional imaging at ultra high magnetic fields: an application to the analysis of resting state networks

    Neuroimage

    (2011)
  • F. De Martino et al.

    Frequency preference and attention effects across cortical depths in the human primary auditory cortex

    Proc. Natl. Acad. Sci. USA

    (2015)
  • F. De Martino et al.

    Spatial organization of frequency preference and selectivity in the human inferior colliculus

    Nat. Commun.

    (2013)
  • F. De Martino et al.

    High-Resolution Mapping of Myeloarchitecture In Vivo: localization of Auditory Areas in the Human Brain

    Cereb. Cortex

    (2014)
  • F. De Martino et al.

    Cortical Depth Dependent Functional Responses in Humans at 7T: improved Specificity with 3D GRASE

    PLoS One

    (2013)
  • Cited by (78)

    View all citing articles on Scopus
    1

    Equal contribution.

    View full text