Kolawole O. Babalola

Computing Accurate Correspondences across Groups of Images

Groupwise image registration algorithms seek to establish dense correspondences between sets of images. Typically, they involve iteratively improving the registration between each image and an evolving mean. A variety of methods have been proposed, which differ in their choice of objective function, representation of deformation field, and optimization methods. Given the complexity of the task, the final accuracy is significantly affected by the choices made for each component. Here, we present a groupwise registration algorithm which can take advantage of the statistics of both the image intensities and the range of shapes across the group to achieve accurate matching. By testing on large sets of images (in both 2D and 3D), we explore the effects of using different image representations and different statistical shape constraints. We demonstrate that careful choice of such representations can lead to significant improvements in overall performance.

Comparison and Evaluation of Segmentation Techniques for Subcortical Structures in Brain MRI

The automation of segmentation of medical images is an active research area. However, there has been criticism of the standard of evaluation of methods. We have comprehensively evaluated four novel methods of automatically segmenting subcortical structures using volumetric, spatial overlap and distance-based measures. Two of the methods are atlas-based - classifier fusion and labelling (CFL) and expectation-maximisation segmentation using a dynamic brain atlas (EMS), and two model-based - profile active appearance models (PAM) and Bayesian appearance models (BAM). Each method was applied to the segmentation of 18 subcortical structures in 270 subjects from a diverse pool varying in age, disease, sex and image acquisition parameters. Our results showed that all four methods perform on par with recently published methods. CFL performed significantly better than the other three methods according to all three classes of metrics.

Using parts and geometry models to initialise Active Appearance Models for automated segmentation of 3D medical images

In recent years, statistical shape models, of which Active Appearance Models (AAMs) are a subset have been increasingly applied to the automatic segmentation of medical images. AAMs are a local search technique requiring good initialisation. In 3D automatic initialisation can be achieved by multiple initialisations, registration, template matching or by application dependent heuristics. The first three can be sub-optimal in certain situations, whilst the last is not generic.

Canonical correlation analysis of sub-cortical brain structures using non-rigid registration

In this paper, we present the application of canonical correlation analysis to investigate how the shapes of different structures within the brain vary statistically relative to each other. Canonical correlation analysis is a multivariate statistical technique which extracts and quantifies correlated behaviour between two sets of vector variables. Firstly, we perform non-rigid image registration of 93 sets of 3D MR images to build sets of surfaces and correspondences for sub-cortical structures in the brain. Canonical correlation analysis is then used to extract and quantify correlated behaviour in the shapes of each pair of surfaces. The results show that correlations are strongest between neighbouring structures and reveal symmetry in the correlation strengths for the left and right sides of the brain.

Comparing the similarity of statistical shape models using the bhattacharya metric

A variety of different methods of finding correspondences across sets of images to build statistical shape models have been proposed, each of which is likely to result in a different model. When dealing with large datasets (particularly in 3D), it is difficult to evaluate the quality of the resulting models. However, if the different methods are successfully modelling the true underlying shape variation, the resulting models should be similar. If two different techniques lead to similar models, it suggests that they are indeed approximating the true shape change. In this paper we explore a method of comparing statistical shape models by evaluating the Bhattacharya overlap between their implied shape distributions. We apply the technique to investigate the similarity of three models of the same 3D dataset constructed using different methods.

A parts-and-geometry initialiser for 3D non-rigid registration using features derived from spin images

Abstract Non-rigid registration is an important precursor to statistical analysis and machine learning in medical image analysis. It is commonly used to find correspondences between images which is a necessary first step for further processing. However, registering images which have large pose differences and/or are composed of substructures of similar appearance requires that registration be initialised carefully for the results to be valid. This work addresses both problems in the context of 3D volumetric images. We use parts-and-geometry models to automatically align images before registration proceeds. An important component of the parts are orientation-invariant descriptors computed using spin images. In the following we describe the construction of the parts-and-geometry models and how they can be incorporated into non-rigid registration. We use 3D CT images of the wrist and knee to demonstrate the effectiveness of the models at locating substructures with similar appearance, and show both qualitatively and quantitatively that initialisation with parts-and-geometry models improve the accuracy of registration.

Registering richly labelled 3D images

We propose a method of registering 3D images in which many regions have been segmented and labelled. Images in which some regions have been labelled can be registered by generating a vector valued image with a number of planes, one for each individual label class, and applying registration algorithms to the multi-plane images. However, when there are many labels such an approach can lead to impractically large images. We demonstrate that good results can be obtained by mapping each label value to a vector in a low dimensional space and applying a multi-plane registration algorithm to the resulting vector image. For the approach to work well, the vectors used for each label should be well separated, and chosen in such a way that there is minimal confusion between them. We demonstrate the method by using it to construct statistical shape models by applying a groupwise alignment method to a set of richly labelled 3D brain images

Lateral asymmetry in the shape of brain ventricles in control and schizophrenia groups

We use point distribution models (PDMs) to investigate lateral asymmetries in the shape of brain ventricles between control subjects and people with schizophrenia. Ventricle surfaces were extracted from T2-weighted MR images and PDMs generated using structural correspondences on the individual surfaces. Using paired linear discriminant analysis we calculate the vector in shape space that maximally separates the shapes of right and left ventricles in the group. The magnitude of the asymmetry is quantified by projection of the individual ventricle shapes onto this vector. We observe significant differences in the magnitude of the asymmetry in both schizophrenia and control groups. There is also a clear difference in the pattern of asymmetry. Male and female subgroups show different magnitudes and patterns of asymmetry, in both groups

Corrigendum to “An evaluation of four automatic methods of segmenting the subcortical structures in the brain” [NeuroImage 47 (2009) 1435–1447]

a University of Manchester, Imaging Science and Biomedical Engineering, Stopford Building, Oxford Road, Manchester M13 9PT, UK b University of Oxford, John Radcliffe Hospital, FMRIB Centre, Oxford, OX3 9DU, UK c Imperial College London, Department of Computing, 180 Queens Gate, London, SW7 2BZ, UK d University of Oxford, Department of Engineering Science, Oxford, OX1 3PJ, UK e MGH/MIT/HMS/, Athinoula A. Martinos Center for Biomedical Imaging, Building 149, 13th Street, Radiology/CNY 149-Room 2301, Charlestown, MA 02129, USA f Institute of Paychiatry, Box PO89, De Crespigny Park, London, SE5 8AF, UK

An Investigation of Morphometric Changes in the Lateral Ventricles of Schizophrenic Subjects

We present results of morphometric analysis of the lateral ventricles of a group of schizophrenic and control subjects to investigate possible shape differences associated with schizophrenia. Our results show shape changes localised to three regions : the temporal horn (its tip near the amygdala, and along its body near the parahippocampal fissure), the central part of the lateral ventricles around the corpus callosum, and the tip of the anterior horn in the region of the frontal lobe. The differences in the temporal and anterior horns are in regions close to structures thought to be implicated in schizophrenia. The changes observed are the most significant changes (p < 10− 13) in shape parameters calculated using a 3D statistical shape descriptor (point distribution model). Corresponding points on the surface of the ventricles in the training set were obtained using an transportation-based method to match high curvature points.