This dissertation proposes a new method of automated malignancy recognition in contrast-enhanced magnetic resonance images of the human breast using the multi-layer perceptron (MLP) feed-forward neural network paradigm. The fundamental limitation is identified as being the efficiency of such a classifier: the computational budget demanded by multi-dimensional image data sets is immense. Without optimization the MLP flounders. This work proposes a new efficient algorithm for MLP classification of large multi-dimensional data sets based on fast discrete orthogonal transforms. This is possible given the straightforward observation that point-wise mask-processing of image data for classification purposes is linear spatial convolution. The novel observation, then, is that the MLP permits convolution at its input layer due to the linearity of the inner product which it computes. Optimized fast Fourier transform (FFT) are investigated and an order of magnitude improvement in the execution time of a four-dimensional transform is achieved over commonly-implemented FFTs. One of the principal retardations in common multi-dimensional FFTs is observed to be the lack of attention paid to memory-hierarchy considerations. A simple, but fast, technique for optimizing cache performance is implemented. The abstract mathematical basis for convolution is investigated and a finite integer number theoretic transform (NTT) approach suggests itself, because such a transform can be defined that is fast, purely real, has parsimony of memory requirements, and has compact hardware realizations. A new method for multi-dimensional convolution with long-length number theoretic transforms is presented. This is an extension of previous work where NTTs were implemented over pseudo-Mersenne, and pseudo-Fermat surrogate moduli. A suitable modulus is identified which allows long-length transforms that readily lend themselves to the multi-dimensional convolution problem involved in classifying large magnetic resonance image data sets.

No description available.

Design for validation is aimed at designing medical devices to make them easier and more economic to validate. Changes to the medical device regulations within the past few years have forced the industry to focus attention on design and validation in order to ensure that a device and its associated manufacturing and test equipment are reliable and fit for purpose. In order for this to happen, design must affect validation and validation must affect design. However, current guidance on validation as it applies to design is inadequate and, as a result, validation is generally not well understood amongst medical device designers. The goal of design for validation is to provide guidance in order to help designers achieve integrated design, development and validation. It forms part of a wider definition of Good Design Practice which aims to encourage fitness for purpose within commercial reality. Exploratory research was carried out in the form of studying ideal practice and current practice in order to identify the factors which contribute to integrated design, development and validation. Case studies were analysed, a model of ideal practice was developed and interviews were carried out with various medical device designers and project managers. From the information gathered, two basic designer needs were identified which had to be fulfilled in order for designers to integrate design, development and validation. A practical approach to design for validation was formulated in order to address the two designer needs through the use of a model of design for validation and a series of six design tactics. The approach was evaluated by sending questionnaires to industry. The feedback was very positive and, based on the evaluation, revisions were made to the design for validation model and design tactics. The revisions will be carried forward to the next phase of the research which is the development of a Design for Validation Workbook.

No description available.

No description available.

Magnetic resonance imaging enables the parameterisation of various physiological processes by comparison of a baseline set of images with a set of images taken after some perturbation of the nuclear spins. By linking this perturbation to a physiological process, changes in signal intensity become a proxy measure of that physiological process. This thesis is concerned with the accuracy and meaning of the measurements made using the natural self-diffusion of water. The first section deals with diffusion tensor measurements, which are three-dimensional measures of water self-diffusion using linear magnetic field gradients. The thesis looks in turn at the effect of signal noise, and partial volume effects on parameters such as the mean diffusivity, the eigenvalues and the fractional anisotropy of the diffusion tensor. The accuracy of the technique, as indicated by the results of numerical simulations, is compared to those acquired from phantom and brain imaging. These results provide an estimate of the dependence of diffusion measurements on technique. The coefficient of variation of diffusion measurements is shown to depend on the metric being used. The voxel volume is shown to have a large effect on the measured anisotropy but not the mean diffusivity. The elevation in diffusion anisotropy observed in acute stroke was then investigated using the eigenvalues of the diffusion tensor, and isotropic and anisotropic tensors. The second section addresses the relationship between gradient-echo and spin-echo dynamic-susceptibility-contrast measurements of cerebral perfusion. The basic contrast measurement is the diffusion of water through pseudo-random magnetic fields created by the injection of a contrast agent. Numerical simulations were used to elucidate the relationship between signal loss and biophysical variables, such as the vessel geometry, the blood volume, and the rate of water self-diffusion.

No description available.

No description available.

No description available.

The aim of my project was to associate breast density with cancer risk by comparing SMF to the conventional density measurement methods. This case control study comprised cancers with age-matched controls from the Cambridge and Norwich Breast Screening Programmes. Data collection involved assessing the films based on Wolfe’s patterns, SCC and 21-categorisation classification, and then digitising the films for computer analyses (Cumulus and SMF). Conditional logistic regression was used to produce odds ratios associated with mammographic density. Results from a pilot study of 220 cancers and 220 controls demonstrated that compared to the conventional methods the SMF measurement was the most effective means of predicting breast cancer risk. This was shown by a methods comparison graph illustrating that the SMF percent volume measurement scheme had the highest density-risk association. Compared to the other density assessment methods, SMF measure of percent volume most clearly discriminated between women at high risk for breast cancer from those at low risk (odds ratio up to 27.1). The study was then expanded to increase the population size to 505 cancers with 1830 controls. Results from the full study contradicted those from the pilot study by showing that SMF measurements were not as effective at predicting breast cancer risk as the gold standard Cumulus measurements. The odds ratios showed a risk of 1.79 (95% CI 1.26-2.55) using SMF percent volume method compared to 3.38 (95% CI 2.32-4.92) as obtained by Cumulus absolute area measurements. I looked into the inconsistencies in the results from the pilot and full studies. After I examined several factors including digitiser effect, reader subjectivity and SMF version effect, I concluded that the promising results in the pilot study were most likely due to chance as I could not explain them by epidemiological or clinical means.