Maja Pokric

The Importance of Partial Voluming in Multi-dimensional Medical Image Segmentation

The presented method addresses the problem of multi-spectral image segmentation through use of a model which takes into account partial volumes of tissues being present in a single voxel at boundaries. The parameters of the multi-dimensional model of pure tissues and their mixtures are iteratively adjusted using an Expectation Maximisation (EM) optimisation technique. Bayes theory is used to generate probability maps for each segmented tissue which estimates the most likely tissue volume fraction within each voxel.

Bayesian and non-Bayesian probabilistic models for medical image analysis

Abstract Bayesian approaches to data analysis are popular in machine vision, and yet the main advantage of Bayes theory, the ability to incorporate prior knowledge in the form of the prior probabilities, may lead to problems in some quantitative tasks. In this paper we demonstrate examples of Bayesian and non-Bayesian techniques from the area of magnetic resonance image (MRI) analysis. Issues raised by these examples are used to illustrate difficulties in Bayesian methods and to motivate an approach based on frequentist methods. We believe this approach to be more suited to quantitative data analysis, and provide a general theory for the use of these methods in learning (Bayes risk) systems and for data fusion. Proofs are given for the more novel aspects of the theory. We conclude with a discussion of the strengths and weaknesses, and the fundamental suitability, of Bayesian and non-Bayesian approaches for MRI analysis in particular, and for machine vision systems in general.

Empirical evaluation of covariance estimates for mutual information coregistration

Mutual information has become a popular similarity measure in multi-modality medical image registration since it was first applied to the problem in 1995. This paper describes a method for calculating the covariance matrix for mutual information coregistration. We derive an expression for the matrix through identification of mutual information with a log-likelihood measure. The validity of this result is then demonstrated through comparison with the results of Monte-Carlo simulations of the coregistration of T1-weighted to T2-weighted synthetic and genuine MRI scans of the brain. We conclude with some observations on the theoretical basis of the mutual information measure as a log-likelihood.

Noise Filtering and Testing Illustrated Using a Multi-Dimensional Partial Volume Model of MR Data

One of the most common problems in image analysis is the estimation and removal of noise or other artefacts using spatial lters. Common techniques include Gaussian, Median and Anisotropic Filtering. Though these techniques are quite common they must be used with great care on medical data, as it is very easy to introduce artifact into images due to spatial smoothing. The use of such techniques is further restricted by the absence of a ‘gold standard’ data against which to test the behaviour of the lters. Following a general discussion of the equivalence of ltering techniques to likelihood based estimation using an assumed model, this paper describes an approach to noise ltering in multi-dimensional data using a partial volume data density model. The resulting data sets can then be taken as a gold standard for spatial ltering techniques which use the information from single images. We demonstrate equivalence between the results from this analysis and techniques for performance characterisation which do not require a ‘gold standard’.

Testing of gadolinium oxy-sulphide phosphors for use in CCD-based X-ray detectors for macromolecular crystallography

The resolution and detective quantum efficiency of CCD-based detectors used for X-ray diffraction is primarily affected by the layer of phosphor that converts incident X-ray photons into visible photons. The optimum thickness of this phosphor layer is strongly dependent on the fraction of absorbed incident X-ray photons and required spatial resolution. A range of terbium doped gadolinium oxy-sulphide (Gd2O2S:Tb) phosphor samples, provided by Applied Scintillation Technologies, have been evaluated for spatial resolution, light output and uniformity. The phosphor samples varied in coating weight (10-25 mg/cm2), grain size (2.5, 4, 10 μm), and applied coating (no coating, reflectors and absorbers). In addition, a non-uniform layer was introduced to some samples in order to provide an inherent diffusion layer. The experimental results showed that the introduction of a reflector increases the point spread function (PSF) and increases light yield up to 30%, while an absorber reduces the PSF tails and decreases the light yield up to 35%. The PSF linearly increases with thickness, while the greatest light yield was obtained with phosphor samples of 4 μm particle size.

Large area high-resolution CCD-based X-ray detector for macromolecular crystallography

An X-ray detector system for macromolecular crystallography based on a large area charge-coupled device (CCD) sensor has been developed as part of a large research and development programme for advanced X-ray sensor technology, funded by industry and the Particle Physics and Astronomy Research Council (PPARC) in the UK. The prototype detector consists of two large area three-sides buttable charge-coupled devices (CCD 46-62 EEV), where the single CCD area is 55.3 mm 41.5 mm. Overall detector imaging area is easily extendable to 85 mm 110 mm. The detector consists of an optically coupled X-ray sensitive phosphor, skewed fibre-optic studs and CCDs. The crystallographic measurement requirements at synchrotron sources are met through a high spatial resolution (2048 1536 pixel array), high dynamic range (B10 5 ), a fast readout (B1 s), low noise (o10e) and much reduced parallax error. Additionally, the prototype detector system has been optimised by increasingits efficiency at low X-ray energies for use at conventional lab sources. The system design of the prototype detector is discussed and the proposed method for crystallographic data processing is briefly outlined. r 2002 Elsevier Science B.V. All rights reserved.

Development of large-area CCD-based x-ray detector for macromolecular crystallography

The design and development of an area CCD-based X-ray detector system, using the first CCD imagers specially designed for macromolecular crystallography, is presented. The system is intended to produce the highest quality data for physically small crystals at synchrotron sources through the use of large CCDs--that is approaching wafer scale. This work is part of a large research and development program for advanced X-ray sensor technology, funded by industry and the Particle Physics and Astronomy Research Council in the UK. The detector has been optimized by increasing its efficiency at low X-ray energies for conventional laboratory sources, and offers fast readout and high dynamic range needed for efficient measurements at synchrotron sources. The detector consists of CCDs optically coupled to a X-ray sensitive phosphor via skewed fiber-optic studs. The individual three- sides buttable CCD consists of 2048 X 1536 27 micrometers square pixels (55.3 X 41.5 mm). The pixel size has been optimized to match diffraction spot profiling needs and the high dynamic range required for such applications. The multiple amplifier outputs possess switched responsivity to maximize the trade-off between signal handling capabilities and linearity. The readout noise is 5 electrons rms at a 1 MHz pixel rate at the high responsivity setting. A prototype detector system comprising two close-butted cooled CCDs is being developed. This system employs a high-efficiency scintillator with very low point spread function, skewed optical-fiber studs (instead of the more usual demagnifying tapers) to maximize the systems detective quantum efficiency and minimize optical distortions. Full system specifications and a novel crystallographic data processing are presented.