Nigel Allinson - Academia.edu (original) (raw)

Papers by Nigel Allinson

In: Proc. of IEEE Int. Workshop on Neural Networks for Signal Processing; IEEE; 2002. p. 525-534., 2002

Advances in Self-Organising Maps, 2001

Catastrophic interference (CI) is a common problem facing many artificial neural network models i... more Catastrophic interference (CI) is a common problem facing many artificial neural network models in learning new data sequentially. Some studies show that the catastrophic interference problem is largely a consequence of the overlap of internal distributed representations, and neural networks with local representations may suffer less. One particular local representation network under study is the Kohonen self-organizing map (SOM).Little is known whether the SOM network may also exhibit this problem.The experiments use focused and cumulative learning with two different sets of data.Results show that SOM may well suffer rom the catastrophic interference problem for some particular selections of parameters and data sets.

Encyclopedia of Biometrics, 2015

PRaVDA Radiotherapy is a fundamental weapon in the battle against cancer with some 40% of patient... more PRaVDA Radiotherapy is a fundamental weapon in the battle against cancer with some 40% of patients receiving it as part of their treatment. Proton therapy (PT) enables a lower integrated radiation dose to a patient receiving radiotherapy (compared to x-rays) due to the finite range of protons and so allows more accurate targeting of the dose. The underlying physics that permits this is the proton's Bragg peak, which increases the dose deposited at a tumour site, even if deep inside the body, as well as reducing the dose to neighbouring healthy tissue. However, PT is more sensitive to uncertainties in both treatment planning and delivery than conventional x-ray treatment . To overcome these limitations of these uncertainities in planning and delivering PT, the Wellcome Trust is funding the PRaVDA Consortium to develop new concepts and instrumentation to provide accurate information about the proton beam’s dose, energy and profile before and during treatment. PRaVDA is a team of l...

Physical Review Applied

The determination of relative stopping power (RSP) via proton computed tomography (pCT) of a pati... more The determination of relative stopping power (RSP) via proton computed tomography (pCT) of a patient is dependent in part on the knowledge of the incoming proton kinetic energies; the uncertainty in these energies is in turn determined by the proton source-typically a cyclotron. Here we show that reducing the incident proton beam energy spread may significantly improve RSP determination in pCT. We demonstrate that the reduction of beam energy spread from the typical 1.0% (at 70 MeV) down to 0.2%, can be achieved at the proton currents needed for imaging at the Paul Scherrer Institut 230 MeV cyclotron. Through a simulated pCT imaging system, we find that this effect results in RSP resolutions as low as 0.2% for materials such as cortical bone, up to 1% for lung tissue. Several materials offer further improvement when the beam (residual) energy is also chosen such that the detection mechanisms used provide the optimal RSP resolution.

Nuclear Instruments and Methods in Physics Research Section A: Accelerators, Spectrometers, Detectors and Associated Equipment, 2016

For many years, silicon micro-strip detectors have been successfully used as tracking detectors f... more For many years, silicon micro-strip detectors have been successfully used as tracking detectors for particle and nuclear physics experiments. A new application of this technology is to the field of particle therapy where radiotherapy is carried out by use of charged particles such as protons or carbon ions. Such a treatment has been shown to have advantages over standard x-ray radiotherapy and as a result of this, many new centres offering particle therapy are currently under construction around the world today. The Proton Radiotherapy, Verification and Dosimetry Applications (PRaVDA) consortium are developing instrumentation for particle therapy based upon technology from high-energy physics. The characteristics of a new silicon micro-strip tracker for particle therapy will be presented. The array uses specifically designed, large area sensors with technology choices that follow closely those taken for the ATLAS experiment at the HL-LHC. These detectors will be arranged into four units each with three layers in an x-u-v configuration to be suitable for fast proton tracking with minimal ambiguities. The sensors will form a tracker capable of tracing the path of $ 200 MeV protons entering and exiting a patient allowing a new mode of imaging known as proton computed tomography (pCT). This will aid the accurate delivery of treatment doses and in addition, the tracker will also be used to monitor the beam profile and total dose delivered during the high fluences used for treatment. We present here details of the design, construction and assembly of one of the four units that will make up the complete tracker along with its characterisation using radiation tests carried out using a 90 Sr source in the laboratory and a 60 MeV proton beam at the Clatterbridge Cancer Centre.

International Symposium on Image and Signal Processing and Analysis, Oct 30, 2009

To enhance the performance of shoeprint recognition systems, an approach capable of extracting th... more To enhance the performance of shoeprint recognition systems, an approach capable of extracting the information-rich 3D outsole patterns is regarded as a promising one. In this paper, initial work on this approach is reported. In this method, 3D outsole models captured using a 3D scanner are sliced in stripes. Stripes are subsequently fitted to parabolas to discover the outsole profiles. Convex/Concave features are hence extracted from each stripe and further fitted by a parametric model to estimate the feature centre position, and the vertical and the horizontal scales. Finally, by grouping estimated features together, a Fuzzy C-Means based method for extracting Printable 3D Features from Convex-Pattern-Dominant Outsoles (Convex-PDOs) is proposed. Promising experimental results show the feasibility of our model-based method for further 3D feature extraction.

Proceedings of the IEEE/IAFE/INFORMS 2000 Conference on Computational Intelligence for Financial Engineering (CIFEr) (Cat. No.00TH8520), 2000

This article considers the dynamics of speculative trading ranges. Daily trading ranges provide g... more This article considers the dynamics of speculative trading ranges. Daily trading ranges provide good estimates of the level of speculative volatility, and analysis of the daily trading range of twenty US futures markets finds that first order differences of the logarithm of daily range show significant negative autocorrelation. This mean-reverting process is also revealed with Hurst analysis. Spectral analysis shows that the underlying dynamics of speculative trading ranges is a pink noise process with each futures market yielding a spectral exponent below that of brown noise.

2006 IEEE Workshop on Signal Processing Systems Design and Implementation, 2006

Fast and efficient operation is a major challenge for complex image processing algorithms execute... more Fast and efficient operation is a major challenge for complex image processing algorithms executed in hardware. This paper describes novel algorithms for correcting optical geometric distortion in imaging systems, together with the architectures used to implement them in FPGA-based hardware. The proposed architecture produces a fast, almost real-time solution for the correction of image distortion implemented using VHDL with a

2006 8th international Conference on Signal Processing, 2006

A novel algorithm for correcting optical spatial distortion in imaging systems is presented, toge... more A novel algorithm for correcting optical spatial distortion in imaging systems is presented, together with the architectures implemented for FPGA-based hardware. Since coordinate transformation functions generally require square root and trigonometric functions, which are not easily calculated using FPGA hardware, the CORDIC (coordinate rotation digital computer) algorithms are employed for these function implementations. An effective pipelined implementation for the radius correction is also demonstrated. The hardware architecture of the spatial correction algorithm has been targeted successfully on a Xilinx Spartan 3 device using the minimum of slices. The experimental results show that the pincushion distortion correction algorithms produce a very low residual error

Electronics Letters, 1998

Fast committee learning can, to some extent, achieve the generalisation advantages of a committee... more Fast committee learning can, to some extent, achieve the generalisation advantages of a committee of neural networks, without the need for independent learning of the committee members. This is achieved by selecting committee members from time-slices of the learning trajectory of one neural network.

Journal of Instrumentation, 2015

Proton radiotherapy has demonstrated benefits in the treatment of certain cancers. Accurate measu... more Proton radiotherapy has demonstrated benefits in the treatment of certain cancers. Accurate measurements of the proton stopping powers in body tissues are required in order to fully optimise the delivery of such treaments. The PRaVDA Consortium is developing a novel, fully solid state device to measure these stopping powers. The PRaVDA Range Telescope (RT), uses a stack of 24 CMOS Active Pixel Sensors (APS) to measure the residual proton energy after the patient. We present here the ability of the CMOS sensors to detect changes in the signal sizes as the proton traverses the RT, compare the results with theory, and discuss the implications of these results on the reconstruction of proton tracks.

Journal of Instrumentation, Jun 3, 2015

Since the first proof of concept in the early 70s, a number of technologies has been proposed to ... more Since the first proof of concept in the early 70s, a number of technologies has been proposed to perform proton CT (pCT), as a means of mapping tissue stopping power for accurate treatment planning in proton therapy. Previous prototypes of energy-range detectors for pCT have been mainly based on the use of scintillator-based calorimeters, to measure proton residual energy after passing through the patient. However, such an approach is limited by the need for only a single proton passing through the energy-range detector in a read-out cycle. A novel approach to this problem could be the use of pixelated detectors, where the independent read-out of each pixel allows to measure simultaneously the residual energy of a number of protons in the same read-out cycle, facilitating a faster and more efficient pCT scan. This paper investigates the suitability of CMOS Active Pixel Sensors (APSs) to track individual protons as they go through a number of CMOS layers, forming an energy-range telescope. Measurements performed at the iThemba Laboratories will be presented and analysed in terms of correlation, to confirm capability of proton tracking for CMOS APSs.

Physics in Medicine and Biology, Jun 9, 2014

Recently CMOS Active Pixels Sensors (APSs) have become a valuable alternative to amorphous Silico... more Recently CMOS Active Pixels Sensors (APSs) have become a valuable alternative to amorphous Silicon and Selenium Flat Panel Imagers (FPIs) in bio-medical imaging applications. CMOS APSs can now be scaled up to the standard 20 cm diameter wafer size by means of a reticle stitching block process. However despite wafer scale CMOS APS being monolithic, sources of non-uniformity of response and regional variations can persist representing a significant challenge for wafer scale sensor response. Nonuniformity of stitched sensors can arise from a number of factors related to the manufacturing process, including variation of amplification, variation between readout components, wafer defects and process variations across the wafer due to manufacturing processes. This paper reports on an investigation into the spatial non-uniformity and regional variations of a wafer scale stitched CMOS APS. For the first time a per-pixel analysis of the electro-optical performance of a wafer CMOS APS is presented, to address inhomogeneity issues arising from the stitching techniques used to manufacture wafer scale sensors. A complete model of the signal generation in the pixel array has been provided and proved capable of accounting for noise and gain variations across the pixel array. This novel analysis leads to readout noise and conversion gain being evaluated at pixel level, stitching block level and in regions of interest, resulting in a coefficient of variation ≤ 1.9%. The uniformity of the image quality performance has been further investigated in a typical X-ray application, i.e. mammography, showing a uniformity in terms of CNR among the highest when compared with mammography detectors commonly used in clinical practise. Finally, in order to compare the CONFIDENTIAL-FOR REVIEW ONLY PMB-100279.R1 detection capability of this novel APS with the currently used technology (i.e. FPIs), theoretical evaluation of the Detection Quantum Efficiency (DQE) at zero-frequency has been performed, resulting in a higher DQE for this detector compared to FPIs. Optical characterization, X-ray contrast measurements and theoretical DQE evaluation suggest that a trade off can be found between the need of a large imaging area and the requirement of a uniform imaging performance, making the DynAMITe large area CMOS APS suitable for a range of bio-medical applications.

Proceedings, May 29, 1995

The phenomenal interest over the last few years in modelling recognition and cognitive processes ... more The phenomenal interest over the last few years in modelling recognition and cognitive processes within a neural network or connectionist framework has resulted in numerous attempts to develop realisations of such systems using optical and VLSI technologies. A neural network is an interconnected structure of many simple nonlinear processing elements which learn from examples to form an internal representation of a problem. Computation is performed collectively by these processing elements, and hence activity is distributed throughout the network. Inherent in this brief description of network operation is the high degree of parallelism present. In classical pattern recognition terms, the feature metrics which make individual object classes similar, result in the formation of clusters in n-dimensional pattern space. For multiple layer perceptrons (MLPs)-the most widely exploited network topology-these clusters are isolated by surrounding each cluster by decision hyperplanes. The MLP is trained on supplied examples of each pattern class, and the decision regions are positioned by some form of gradient descent algorithm which iteratively adapts the synaptic weights of each neuron. MLPs are an example of supervised learning in a feedforward network. The authors concentrate on a purely digital realisation of neural networks based on an unsupervised learning situation, which is a form of adaptation to an unknown environment. >

International Journal of Electronics, Dec 1, 1985

A detailed analysis of the distortion-reducing properties of current-dumping amplifiers is presen... more A detailed analysis of the distortion-reducing properties of current-dumping amplifiers is presented, which takes into account the finite tolerance of components. The amplifier is shown to possess distortion-reduction capabilities superior to those of conventional feedback amplifiers, especially at high frequencies. It is an example of a class of amplifiers that can be termed frequency-dependent feedback-feedforward amplifiers.