Showing papers by "Lionel Tarassenko published in 1995"

Novelty detection for the identification of masses in mammograms

Abstract: Breast cancer is the major cause of death amongst women in the 35 to 55 age group. Mammography is the only feasible imaging modality for screening large numbers of women. With the present screening policy, there are three million mammograms to be analysed each year in the UK; there is therefore a need (as yet unmet) for an automated analysis system which could highlight areas of interest. In the first instance, the areas of interest might simply be any mass-like structures and this is indeed the approach reported on in this paper. Mammography is typical of many problems in medicine: the class of real interest is under-represented in the database of available examples and hence its prior probability will be very low. As a result of this, there are very few examples of abnormalities in any of the existing databases. If a neural network classifier is trained using the standard approach of minimising the mean-squared error (MSE) at the output, the under-represented class will be ignored. We have been exploring an alternative approach in which we attempt to learn a description of normality using the large number of available mammograms which do not show any evidence of mass-like structures. The idea is then to test for novelty against this description in order to try and identify candidate masses in previously unseen images analysis and interpretation and present a sample of the results which we have so far obtained on a standard database.

Novelty detection for the identification of masses in mammograms

Abstract: Breast cancer is the major cause of death amongst women in the 35 to 55 age group. Mammography is the only feasible imaging modality for screening large numbers of women. With the present screening policy, there are three million mammograms to be analysed each year in the UK; there is therefore a need (as yet unmet) for an automated analysis system which could highlight areas of interest. In the first instance, the areas of interest might simply be any mass-like structures and this is indeed the approach reported on in this paper. Mammography is typical of many problems in medicine: the class of real interest is under-represented in the database of available examples and hence its prior probability will be very low. As a result of this, there are very few examples of abnormalities in any of the existing databases. If a neural network classifier is trained using the standard approach of minimising the mean-squared error (MSE) at the output, the under-represented class will be ignored. We have been exploring an alternative approach in which we attempt to learn a description of normality using the large number of available mammograms which do not show any evidence of mass-like structures. The idea is then to test for novelty against this description in order to try and identify candidate masses in previously unseen images analysis and interpretation and present a sample of the results which we have so far obtained on a standard database.

Text-independent speaker recognition using neural network techniques

Abstract: The domain of speaker recognition tasks extends from access control to office automation (such as labelling voice mail, or choosing the correct speaker dependent model for a speech recognition system) to forensics. Despite the range of potential applications for identifying or verifying speakers, there are relatively few current commercial systems because accurate speaker recognition in real world conditions remains a challenging problem. The paper considers the application of neural network techniques to the task of speaker identification. It concentrates on modular extensions to the well known multilayer perceptron (MLP) and radial basis function (RBF) architectures.

Automated Sleep EEg Analysis using an RBF Network

Abstract: There are many examples of expert systems which have been developed in the last twenty years in an attempt to solve medical diagnostic problems automatically (see, for example [1]). There are, however, a number of medical problems which do not lend themselves very well to the expert system’s approach. In this chapter, we focus on one such problem, namely the analysis of the electroencephalogram (EEG) during sleep. At present, a set of rules proposed more than twenty years ago [15] is still being used by human experts to classify successive 30-second segments of the EEG sleep record into one of six major categories (wake, dreaming sleep and four stages of progressively deeper sleep) but the rules are notoriously difficult to apply and inter-observer correlation can be as low as 51% for some sections of data [8]. The lack of agreement amongst trained human experts on all but very typical data segments has made the automation of the “sleep scoring” process an almost impossible task.

Implementation issues for on-chip learning with analogue VLSI MLPS

Abstract: Microelectronic neural network technology has become sufficiently mature over the past few years that reliable performance can now be obtained from VLSI circuits under carefully controlled conditions. The use of analogue VLSI allows low power, low cost and area efficient hardware realisations which can perform the computationally intensive feed-forward operation of neural networks at high speed. These factors, coupled with the ability to interface directly with the analogue world, make real-time applications a possibility. This paper attempts to address some of the issues concerning in-situ learning with analogue VLSI multilayer perceptron (MLP) networks. We consider the modes used to train analogue neural networks, study weight storage and circuit precision issues, and identify the most promising training algorithms. We then make some conclusions based on results from analogue VLSI chips that we have designed, built and successfully tested.

Retinal pulse oximetry: towards a method for measuring cerebral oxygen saturation

Abstract: An instrument to make reflectance pulse oximetry measurements on the blood in the retinal fundus has been developed. Conventional pulse oximeters measure the oxygen saturation of arterial blood in the peripheral circulation, whereas using the retina provides the opportunity to monitor cerebral oxygenation. Cardiac-synchronous pulsations have been observed in the red and IR signals during preliminary in vivo trials, however it is unclear whether these pulsations are entirely due to the pulsating volume of arterial blood. Analysis of a series of retinal images has revealed that pulsations can be detected from the veins and choroidal background as well as from the arteries.

The use of neural networks in EEG analysis

Abstract: There is little doubt that, if appropriately used, artificial `neural' networks (ANNs) offer a robust method for estimation, prediction and classification. Their application to EEG analysis is well-founded. Care must, however, be taken in the choice of pre- and post-processors. Although the use of ANNs does not offer a solution to some of the problems encountered in EEG analysis, the authors would argue that used as methods for providing a continuous measure of a system's state, in a probabilistic framework, they provide information which is lost in more traditional analysis methods. (3 pages)

A design for a visual motion transducer

Abstract: Autonomous vehicles could benefit greatly from visual-motion sensors of sufficiently low cost that 10 or 20 of them could be distributed around the vehicle's periphery. A design is outlined here that builds on well-known schemes using gratings. The grating principle is illustrated by the fact that a man with a torch, walking at night behind a railing, seems to flash. The frequency of flashing is proportional to his velocity. A major drawback is that backward and forward motion are not distinguished. The key development here is the use of commutation as a means of modulating the grating output signal. This is equivalent, in the illustration above, to simulating stroboscopic motion of the railing. Thus, when the man is stationary, there is flashing at a resting frequency. When he moves one way the frequency increases, and for motion the opposite way, frequency decreases. Results from an analogue implementation of the visual motion transducer are presented. The current transducer measures translational motion across the grating. The design is also shown to be capable of extension to direct measurement of the divergence and curl of the flow field. >