Richard Dybowski

Clinical Applications of Artificial Neural Networks

List of contributors 1. Introduction Richard Dybowski and Vanya Gant Part I. Applications: 2. Artificial neural networks in laboratory medicine Simon S. Cross 3. Using artificial neural networks to screen cervical smears: how new technology enhances health care Mathilde E. Boon and Lambrecht P. Kok 4. Neural network analysis of sleep disorders Lionel Tarassenko, Mayela Zamora and James Pardey 5. Artificial neural networks for neonatal intensive care Emma A. Braithwaite, Jimmy Dripps, Andrew J. Lyon and Alan Murray 6. Artificial neural networks in urology: applications, feature extraction and user implementations Craig S. Niederberger and Richard M. Golden 7. Artificial neural networks as a tool for whole organism fingerprinting in bacterial taxonomy Royston Goodacre Part II. Prospects: 8. Recent advances in EEG signal analysis and classification Charles W. Anderson and David A. Peterson 9. Adaptive resonance theory: a foundation for apprentice systems in clinical decision support? Robert F. Harrison, Simon S. Cross, R. Lee Kennedy, Chee Peng Lim and Joseph Downs 10. Evolving artificial neural networks V. William Porto and David B. Fogel Part III. Theory: 11. Neural networks as statistical methods in survival analysis Brian D. Ripley and Ruth M. Ripley 12. A review of techniques for extracting rules from trained artificial neural networks Robert Andrews, Alan B. Tickle and Joachim Diederich 13. Confidence intervals and prediction intervals for feedforward neural networks Richard Dybowski and Stephen J. Roberts Part IV. Ethics and Clinical Prospects: 14. Artificial neural networks: practical considerations for clinical application Vanya Gant, Susan Rodway and Jeremy Wyatt Index.

Clinical applications of artificial neural networks: Confidence intervals and prediction intervals for feedforward neural networks

The chapter opens with an introduction to regression and its implementation within the maximum-likelihood framework. This is followed by a general introduction to classical confidence intervals and prediction intervals. We set the scene by first considering confidence and prediction intervals based on univariate samples, and then we progress to regarding these intervals in the context of linear regression and logistic regression. Since a feed-forward neural network is a type of regression model, the concepts of confidence and prediction intervals are applicable to these networks, and we look at several techniques for doing this via maximum-likelihood estimation. An alternative nto the maximum-likelihood framework is Bayesian statistics, and we examine the notions of Bayesian confidence and predictions intervals as applied to feed-forward networks. This includes a critique on Bayesian confidence intervals and classification.

Classification of incomplete feature vectors by radial basis function networks

The paper describes the use of radial basis function neural networks with Gaussian basis functions to classify incomplete feature vectors. The method uses the fact that any marginal distribution of a Gaussian distribution can be determined from the mean vector and covariance matrix of the joint distribution.

The Effects of Vaccination and Immunity on Bacterial Infection Dynamics In Vivo

Salmonella enterica infections are a significant global health issue, and development of vaccines against these bacteria requires an improved understanding of how vaccination affects the growth and spread of the bacteria within the host. We have combined in vivo tracking of molecularly tagged bacterial subpopulations with mathematical modelling to gain a novel insight into how different classes of vaccines and branches of the immune response protect against secondary Salmonella enterica infections of the mouse. We have found that a live Salmonella vaccine significantly reduced bacteraemia during a secondary challenge and restrained inter-organ spread of the bacteria in the systemic organs. Further, fitting mechanistic models to the data indicated that live vaccine immunisation enhanced both the bacterial killing in the very early stages of the infection and bacteriostatic control over the first day post-challenge. T-cell immunity induced by this vaccine is not necessary for the enhanced bacteriostasis but is required for subsequent bactericidal clearance of Salmonella in the blood and tissues. Conversely, a non-living vaccine while able to enhance initial blood clearance and killing of virulent secondary challenge bacteria, was unable to alter the subsequent bacterial growth rate in the systemic organs, did not prevent the resurgence of extensive bacteraemia and failed to control the spread of the bacteria in the body.

Characteristics of rhegmatogenous retinal detachment in pseudophakic and phakic eyes

AimsTo investigate whether pseudophakic and phakic rhegmatogenous retinal detachment (RRD) patterns differ.MethodsRetrospective review of electronic database of patients, aged 50 years or over, presenting to our vitreoretinal service. Data included baseline characteristics, digital drawings, and outcomes. Retinal drawings were analysed in a masked fashion for site, size, and number of retinal breaks. Comparisons were made between the following groups and subgroups: pseudophakic eyes, phakic eyes, phakic eyes with cataract, and phakic eyes without cataract.ResultsOf 500 eyes included, 146 were pseudophakic; 177 of the phakic eyes had cataract. The following were significant by univariate analysis: pseudophakic patients were older than phakic patients in general, but the same age as patients with cataract; in the pseudophakic group, there were lower proportions of females and of patients presenting with vitreous haemorrhage or with large or superotemporal breaks; higher proportions of pseudophakic eyes had small breaks and inferonasal breaks. Some differences remained significant when comparing pseudophakia eyes with cataract. Multivariate analysis comparing pseudophakia with phakia confirmed a lower chance in pseudophakia of large breaks, vitreous haemorrhage and superotemporal breaks, but higher chance of detached inferior breaks. Some variables were age dependent.ConclusionDifferences were found between pseudophakic and phakic RRD patterns. These suggest special pathogenetic mechanisms in pseudophakic retinal detachment, which could help explain increased incidences of RRD after cataract surgery.

Rapid compound pattern classification by recursive partitioning of feature space. An application in flow cytometry

A method is described for rapidly classifying a set of points in real space. A set is mapped to a low-dimensional vector via a discriminating, recursive partition of feature space obtained pragmatically by the CART algorithm.

Single passage in mouse organs enhances the survival and spread of Salmonella enterica.

Intravenous inoculation of Salmonella enterica serovar Typhimurium into mice is a prime experimental model of invasive salmonellosis. The use of wild-type isogenic tagged strains (WITS) in this system has revealed that bacteria undergo independent bottlenecks in the liver and spleen before establishing a systemic infection. We recently showed that those bacteria that survived the bottleneck exhibited enhanced growth when transferred to naive mice. In this study, we set out to disentangle the components of this in vivo adaptation by inoculating mice with WITS grown either in vitro or in vivo. We developed an original method to estimate the replication and killing rates of bacteria from experimental data, which involved solving the probability-generating function of a non-homogeneous birth–death–immigration process. This revealed a low initial mortality in bacteria obtained from a donor animal. Next, an analysis of WITS distributions in the livers and spleens of recipient animals indicated that in vivo-passaged bacteria started spreading between organs earlier than in vitro-grown bacteria. These results further our understanding of the influence of passage in a host on the fitness and virulence of Salmonella enterica and represent an advance in the power of investigation on the patterns and mechanisms of host–pathogen interactions.

Nested Sampling for Bayesian Model Comparison in the Context of Salmonella Disease Dynamics

Understanding the mechanisms underlying the observed dynamics of complex biological systems requires the statistical assessment and comparison of multiple alternative models. Although this has traditionally been done using maximum likelihood-based methods such as Akaikes Information Criterion (AIC), Bayesian methods have gained in popularity because they provide more informative output in the form of posterior probability distributions. However, comparison between multiple models in a Bayesian framework is made difficult by the computational cost of numerical integration over large parameter spaces. A new, efficient method for the computation of posterior probabilities has recently been proposed and applied to complex problems from the physical sciences. Here we demonstrate how nested sampling can be used for inference and model comparison in biological sciences. We present a reanalysis of data from experimental infection of mice with Salmonella enterica showing the distribution of bacteria in liver cells. In addition to confirming the main finding of the original analysis, which relied on AIC, our approach provides: (a) integration across the parameter space, (b) estimation of the posterior parameter distributions (with visualisations of parameter correlations), and (c) estimation of the posterior predictive distributions for goodness-of-fit assessments of the models. The goodness-of-fit results suggest that alternative mechanistic models and a relaxation of the quasi-stationary assumption should be considered.

An Anthology of Probabilistic Models for Medical Informatics

We present a collection of examples that illustrate how probabilistic models can be applied within medical informatics, along with the relevant statistical theory.

An efficient moments-based inference method for within-host bacterial infection dynamics.

Over the last ten years, isogenic tagging (IT) has revolutionised the study of bacterial infection dynamics in laboratory animal models. However, quantitative analysis of IT data has been hindered by the piecemeal development of relevant statistical models. The most promising approach relies on stochastic Markovian models of bacterial population dynamics within and among organs. Here we present an efficient numerical method to fit such stochastic dynamic models to in vivo experimental IT data. A common approach to statistical inference with stochastic dynamic models relies on producing large numbers of simulations, but this remains a slow and inefficient method for all but simple problems, especially when tracking bacteria in multiple locations simultaneously. Instead, we derive and solve the systems of ordinary differential equations for the two lower-order moments of the stochastic variables (mean, variance and covariance). For any given model structure, and assuming linear dynamic rates, we demonstrate how the model parameters can be efficiently and accurately estimated by divergence minimisation. We then apply our method to an experimental dataset and compare the estimates and goodness-of-fit to those obtained by maximum likelihood estimation. While both sets of parameter estimates had overlapping confidence regions, the new method produced lower values for the division and death rates of bacteria: these improved the goodness-of-fit at the second time point at the expense of that of the first time point. This flexible framework can easily be applied to a range of experimental systems. Its computational efficiency paves the way for model comparison and optimal experimental design.