Staff/Affiliates:

Leigh Johnston, Iven Mareels, David Grayden, Gary Egan, Maria Gavrilescu, 

Students:

Collaborators/Sponsors:

Functional Magnetic Resonance Imaging (fMRI) provides an indirect measure of neuronal activity. The neuronal response to a stimulus in a particular brain region elicits a hemodynamic response in the surrounding capillary networks, due to increased demand for oxygenated blood. The resultant interactions between cerebral blood flow, volume and metabolic rate of oxygen cause local MR signal perturbations, termed the Blood Oxygenation Level Dependent (BOLD) effect. We are interested in the formulation of mathematical models that describe the BOLD effect, and the analysis of these models for the interpretation of fMRI experimental results. The typically employed linear hemodynamic response model is unable to take into account the marked variability in response shape known to exist across cortical regions and between individuals [1]. We aim to develop biologically meaningful nonlinear models of the BOLD response and are applying statistical signal processing techniques for the inference of hidden physiological variables [2]. A second focus of this project is the development of rigorous and reliable methods for estimating connectivity between brain regions as detectable from fMRI experiments. This research advances fundamental understanding of brain function, and is applicable in the development of fMRI-based cognitive neuroscience and pre-surgical planning tools. 

Fig. 3: Observed BOLD signal (right: black) in the primary motor cortex (left: red square). Particle filter estimates of BOLD signal (red) and normalised cerebral blood flow (green), volume (blue) and deoxyhemoglobin content (purple), for a) optimal, b) underfitting and c) over-fitting of system parameters.

[1] E. Duff, J. Xiong, B. Wang, R. Cunnington, P. Fox and G. Egan, "Complex spatio-temporal dynamics of fMRI BOLD: A study of motor learning", NeuroImage, 34, pp.156-168, 2007.

[2] L. Johnston, E. Duff and G. Egan, "Particle filtering for nonlinear BOLD signal analysis", 9th International Conference on Medical Image Computing and Computer Assisted Intervention, 2, pp. 292-299, 2006.

Staff/Affiliates:

Leigh Johnston, Gary Egan, Iven Mareels

Students:

Collaborators/Sponsors:

Magnetic Resonance Imaging is a non-invasive technique of vast neuroscientific benefit, owing to its ability to image the internal structure of the brain. We propose the application of signal processing techniques for improvement in MR signal acquisition, contrast enhancement in the reconstructed image volumes, and development of robust image processing methods, motivated by potential impact on both neuroscience research endeavours and improved clinical and public health outcomes.

Increasingly higher field strength MRI scanners are permitting detection of more detailed brain structures, for example via cortical parcellation algorithms validated on histology datasets (Fig.4)  [1]. Similarly, recent modalities like diffusion MRI are rapidly advancing the ability to noninvasively  study brain structure. Diffusion MRI is sensitive to the directional diffusivity of water, detected via application of magnetic field gradients. White matter fibres, comprised of myelinated axon bundles, are now identifiable in both location and direction. We are developing diffusion MRI analysis methods and tractography algorithms for use in characterisation, and ultimately early detection, of neurological diseases such as Multiple Sclerosis and Huntington's disease.

Fig. 4: Automated parcellation of a post-mortem histological slice of baboon cortex. Left: Haematoxylin & eosin stained slice. Right: Flattened segment of cortex (top), Map of posterior probability of dark band (middle), Cortical parcellation result (bottom). 

Fig. 5: Visualisation of local white matter structure as determined by water diffusivity, in 32-direction diffusion MR image of a Huntington's disease patient.