Dynamically adaptive imaging with real-time MRI

In the last two years we have been developing a new way to use an MRI scanner to investigate how the brain functions. Rather than analyzing the images of brain activity some time after the scan finishes, in this new method the data from the MRI scanner are analyzed in real time and used to dynamically adapt the stimuli or task presented to the volunteer. Effectively, the experiment automatically adapts to the particular volunteer in the scanner, so allowing a much more detailed characterization of their neural representations. This method has many exciting applications in neuroscientific research and in the clinic. To date, we have applied it in healthy participants, to understand their representations of visual and auditory objects, and to examine visual imagery.

Relevant references

Cusack, R. , Veldsman, M. , Naci, L. , Mitchell, D. J. , Linke, A. C. (2011) Seeing Different Objects in Different Ways: Measuring Ventral Visual Tuning to Sensory and Semantic Features With Dynamically Adaptive Imaging. Human Brain Mapping. pdf

Multi-voxel pattern analysis

The principle behind traditional neuroimaging analysis is to identify fairly large (>1-2 cm) regions of the brain become activated as a bulk.  The data are spatially smoothed, which averages out local noise, and improves sensitivity for larger patches of activation. An important development in neuroimaging analysis over the last decade has been to realize that in addition to this bulk signal, the voxel-to-voxel variation in these activation patterns actually carries important information. This is what is exploited in the technique of MVPA.

MVPA allows the discrimination of much more finely distinguished mental states from their evoked neural activity than has previously been possible. Our laboratory has been involved in work to understand the source of the information that is present on a voxel-to-voxel scale, and how to ensure the MRI acquisition is maximally sensitive to it.

MVPA techniques often have much greater power than traditional analysis methods is that because they typically examine the reliability of the relationship between the pattern of neural activity and the mental state within each volunteer. They don’t assume that each volunteer will have the same pattern, which means they remain sensitive even when the neural code differs from person to person, or when inter-subject registration is poor.

Relevant references

Cusack, R. , Veldsman, M. , Naci, L. , Mitchell, D. J. , Linke, A. C. (2011) Seeing Different Objects in Different Ways: Measuring Ventral Visual Tuning to Sensory and Semantic Features With Dynamically Adaptive Imaging. Human Brain Mapping. pdf

Thompson, R. , Correia, M. , Cusack, R. (2010) Vascular contributions to pattern analysis: Comparing gradient and spin echo fMRI at 3T. NeuroImage, Elsevier Inc. pubmed, pdf

Kriegeskorte, N. , Cusack, R. , Bandettini, P. (2009) How does an fMRI voxel sample the neuronal activity pattern: Compact-kernel or complex spatiotemporal filter? NeuroImage. pubmed, pdf

Stokes, M. , Thompson, R. , Cusack, R. , Duncan, J. (2009) Top-down activation of shape-specific population codes in visual cortex during mental imagery. The Journal of Neuroscience 29(5), p. 1565-72. pubmed, pdf

Using phase information

The MRI signal arrives as a complex number, containing both magnitude and phase. Typically, the phase is discarded and only magnitude images are analyzed. However, some MR sequences, such as those that measure small inhomogeneities in the magnetic field across the person’s head or blood flow, are designed to yield information in the phase of the signal. Other MR sequences (such as EPI) also “accidentally” contain useful information in the phase signal.

A challenge in using phase information is that it has a discontinuous and non-monotonic relationship to the thing it measures, due to “phase wrapping” – the fact that we can only observe the final angle of the spins in a 0 to 2 range, and not how many whole rotations they’ve taken to get there. Recovering this signal absolutely is impossible from a single image, but the relative phase of different voxels in an image can be recovered through a process known as phase unwrapping. Together with collaborators we have developed new noise-robust ways of performing phase unwrapping. We have also applied these to the correct of distortion in fMRI images due to magnetic  field inhomogeneities.

Relevant references

Salfity, M. F. , Ruiz, P. D. , Huntley, J. M. , Graves, M. J. , Cusack, R. , Beauregard, D. A.(2006) Branch cut surface placement for unwrapping of undersampled three-dimensional phase data: application to magnetic resonance imaging arterial flow mapping, Applied Optics 45(12), p. 2711-2722, pubmed

Cusack, R. , Russell, B. , Cox, S. M. L. , De Panfilis, C. , Schwarzbauer, C. , Ansorge, R.(2005) An evaluation of the use of passive shimming to improve frontal sensitivity in fMRI. NeuroImage 24(1), p. 82-91, pubmed, pdf

Salfity, M. F. , Huntley, J. M. , Graves, M. J. , Marklund, O. , Cusack, R. , Beauregard, D. A.(2004) 3-D and 4-D phase unwrapping methods applied to phase contrast magnetic resonance velocity imaging, Methods, p. 3-8, pdf

Cusack, R. (2003) An Evaluation of the Use of Magnetic Field Maps to Undistort Echo-Planar Images, NeuroImage 18(1), p. 127-142, url, pdf

Cusack, R. , Papadakis, N. (2002) New Robust 3-D Phase Unwrapping Algorithms: Application to Magnetic Field Mapping and Undistorting Echoplanar Images*1, NeuroImage 16(3), p. 754-764, url, pdf

Cusack, R. , Huntley, J. M. , Goldrein, H. T. (1995) Improved noise-immune phase-unwrapping algorithm, Applied Optics 34(5), p. 781-789, url, url