Neuroimaging and Neuroinformatics

Search

• Interoception
• MR Methods
• Neurological Disorders
• Neuroinformatics
• Animal MRI

Automated Segmentation of Mouse Brain MRI

Mouse models of neurological disease are vital tools in neuroscience research owing to their ability to replicate genetic and phenotypic aspects of human neurological disorders.  MR imaging is increasingly being exploited in the study of brain structure, offering a non-invasive and three-dimensional image acquisition technique in contrast to traditional two-dimensional histology sectioning.

This project concerns the development of automated image analysis techniques for a variety of problems related to characterisation of mouse brain imagery.  The most fundamental of these is the creation of MRI brain atlases of given strains of mouse.  These will enhance existing histological atlases through the non-trivial task of registration of MR and histological data.   A direct consequence of atlas creation will be the development of automated tools for obtaining robust and consistent quantitative measures of structural changes between a mouse model and background strain, both based on image intensity changes and morphometric shape modelling of local brain structures.

The outcome of this project is a significant advance in the availability of automated computational tools for the experimental neuroscientist, and the development of image processing algorithms applicable to any field of science or engineering research concerned with the quantitative analysis and characterisation of image datasets.

Regularisation of High Angular Resolution Diffusion Tensor Imaging Data

The standard Diffusion Tensor Imaging model describes an ellipsoid at each voxel, representing the preferential diffusion of water molecules in that area.  This model provides indirect information about the characteristics of constituent white matter fibre bundles,  but is insufficient for describing more complicated fibre arrangements such as crossing or kissing fibres.   To overcome this shortcoming, attention has turned to High Angular Resolution Diffusion Tensor Imaging (HARDI), in which gradients are applied in a large number of directions.  HARDI data providing more information from which to infer the underlying fibre structure within each voxel.  We have developed a method for noise suppression in high angular resolution diffusion MRI data through direct regularisation of the Apparent Diffusion Coefficient profiles.  The technique is derived in a Bayesian framework in the style of the seminal techniques for image restoration using Markov random field models.  A notable extension to traditional approaches is made in operating over the four dimensional image and diffusion gradient direction space.  

Nonlinear filtering methods for analysis of the BOLD signal in fMRI

The two global aims of this work are to improve the modelling of the BOLD signal formation, and to develop robust estimation methods for doing inference on the more complicated signal models resulting from the first part of the project.  In formulating enhanced BOLD signal models, we aim to model all constituent physiological processes and characterise regional differences throughout the human brain, through Arterial Spin Labelling and fMRI experiments.   We have applied particle filtering, a sequential importance resampling technique, to the problem of inferring hidden physiological state variables from measured BOLD data.   This nonlinear filtering method improves on methods proposed in the literature, and traditional approaches such as the Extended Kalman filter.

Available Student Projects:
Model-based MRI Processing

Wavelet bootstrapping estimation of fMRI data