Uses, misuses, new uses and fundamental limitations of magnetic resonance imaging in cognitive science

Abstract

When blood oxygenation level-dependent (BOLD) contrast functional magnetic resonance imaging (fMRI) was discovered in the early 1990s, it provoked an explosion of interest in exploring human cognition, using brain mapping techniques based on MRI. Standards for data acquisition and analysis were rapidly put in place, in order to assist comparison of results across laboratories. Recently, MRI data acquisition capabilities have improved dramatically, inviting a rethink of strategies for relating functional brain activity at the systems level with its neuronal substrates and functional connections. This paper reviews the established capabilities of BOLD contrast fMRI, the perceived weaknesses of major methods of analysis, and current results that may provide insights into improved brain modelling. These results have inspired the use of in vivo myeloarchitecture for localizing brain activity, individual subject analysis without spatial smoothing and mapping of changes in cerebral blood volume instead of BOLD activation changes. The apparent fundamental limitations of all methods based on nuclear magnetic resonance are also discussed.This article is part of the themed issue 'Interpreting BOLD: a dialogue between cognitive and cellular neuroscience'.

Keywords: brain function; cerebral blood volume; cortical layers; magnetic resonance imaging; myeloarchitecture; neuroanatomy.

Figure 1.

Single subject axial section images acquired with echo planar imaging slices, with colour-overlaid BOLD activation maps during finger movement and imagined finger movement (figure courtesy of Robert Trampel). Spatial resolution 0.75 mm isotropic. (a) Axial section acquired using zoomed EPI shows raw data quality. The yellow line indicates the central sulcus, with ‘hand knob’. (b) Activation map during ‘tapping’ versus ‘rest’. (c) ‘Moving’ versus ‘rest’. (d) Contrast of ‘tapping’ versus ‘moving’. (e) Activation map during imagined finger movement. Colour bars indicate z-scores. Functional maps thresholded at p < 0.05, using false discovery rate.

Figure 2.

Layer-dependence of BOLD activation for finger tapping, finger movement and imagined finger tapping. Grand average time course and mean BOLD signal of nine subjects at four different cortical depths in primary motor cortex. (figure courtesy of Robert Trampel). (a) Grand average BOLD time courses obtained at four cortical depths in primary motor cortex averaged across nine subjects. Error bars represent standard error of the mean. (b) Corresponding mean BOLD signal difference between the three conditions (‘imagining’, ‘tapping’, ‘moving’) and ‘rest’, respectively. Error bars represent standard error of the mean. The dashed circle emphasizes the smaller difference between ‘middle lamina 2’ and ‘deep lamina’ for motor imagery compared to actual motor performance.

Figure 3.

Functional MRI using BOLD contrast and cerebral blood volume maps acquired with slab-selective vascular occupancy imaging (VASO). Comparison of single-subject activation maps generated by finger tapping, in an axial section through the central sulcus. Echo-planar data acquisition, resolution 0.74 × 0.74 × 2 mm³, TE = 20 ms, TR = 1.5 s (interleaved VASO and BOLD acquisition; figure courtesy of Laurentius Huber).