Bumper year for the Statistics and Neural Computation Research Group

You can find PDF copies of the many research papers produced by the group on our website:


A novel method to extract rules from data has been published by the IEEE transactions on neural networks. It is claimed to be the fastest and most accurate method to describe data using low-order Boolean rules and it is now being used within the activities of the European Network of Excellence BioPattern (www.biopattern.org) for a range of medical applications including cancer prognosis and diagnosis.

The method also features in a prototype decision support system for breast cancer. A longstanding collaboration with Alfredo Vellido, now in Barcelona, continues to generate theoretical papers on neural networks. Dr Terence Etchells also collaborated with Alfredo Vellido on a number of projects involving rule extraction from large multidimensional data sets, leading to a number of conference publications and a book chapter in Lecture Notes in Computer Science.

Following on the tracks of a well-cited review of neural network applications in medicine, Professor Paulo Lisboa has teamed-up with a colleague in the Clinical Engineering Department of the Royal Liverpool Hospital & University of Liverpool for an update review on specifically on cancer. More importantly, the prognostic model developed by the group, which is called PLANN-ARD (Partial Logistic Artificial Neural Network with Automatic Relevance Determination) came top in a double-blind evaluation of predictors of survival from eye cancer (uveal melanoma). Finally, joint work with the Clinical Engineering Department and our old colleague Wael El-Deredy showed that Independent Components Analysis can be successfully used to detect the evoked eye response in adults and children. This work has been published in a journal but a related conference paper won the Best Paper Prize at the International Conference MEDSIP (Medical Signal and Image Processing) in Glasgow.

Collaboration also happens within the University’s umbrella Centre for Health and Social Care Informatics (http://www.ljmu.ac.uk/chasci/), this time involving the Research Institute for Sports Science in a visualization system for gait analysis using the Self-Organising Map.

An engineering collaboration with LJMU’s General Engineering Research Institute (GERI) recently led to a paper on facial image recognition. A special feature of this method is that it is based on a computational model of the human visual system, so that faces are recognized by identifying key features such as the eyes, nose and mouth. As a result, facial feature understanding comes form free with this model.

Example of the groups work: a cluster of malignant patients

discovered by rule extraction and visualized in 3-D, from a high

dimension cancer data set.

Growing collaboration within the School is also resulting in published joint work. A joint paper has been published on financial time-series analysis, continuing a tradition started with an earlier work with the Centre for International Banking Economics & Finance. For more details, see:


The group started this year an EPSRC-CASE studentship funding Inigo Nafarrate Zubizarreta. This grant has part-funding from Unilever Corporate Research and was brokered by the Smith Institute for Mathematics, from Oxford. The 3-year studentship is to develop Bayesian Graphical Models for large-scale Personalised Recommender systems. However, this work links closely with the Biopattern research as the same method can, in principle, be used to understand bioinformatics data for medical applications. More on this later in the year!

This year found Terence collaborating with New Zealand Anaesthetist Prof Michael Harrison using the OSRE methodology to find rules for the decision to perform a blood transfusion during surgery leading to a paper in the medical journal Anaesthesia. Further collaboration led to a paper being presented at the New Zealand Anaesthesia Annual Scientific Meeting on the determination of the mode of ventilation used during surgery, discovered from data using OSRE.

Terence was also invited to give a keynote lecture at the DES-TIME conference in Berlin July 2006. DESTIME is a well established biennial conference on the use of computer algebra in mathematics teaching. The title of Terence’s Lecture was ‘Generating Online Assessment Questions with Derive and Perception’.

The summer saw the start of an ongoing consultancy with Themis, a private contract research company specialising in clinical studies. Ian Jarman and Hane Aung’s research compared a standard multivariate model for censored data (Cox regression) with a Partial Logistic Artificial Neural Network (PLANN) regularised with Automatic Relevance Determination (ARD), applied to a time-to-event data set of baseline clinical and histological indicators for breast cancer prognosis looking at mortality and treatment failure. This work has been extended into competing risks models for local and distal recurrence.

Ian Jarman successfully defended his PhD thesis, “An integrated framework for risk profiling of breast cancer patients following surgery”. This involved the development of the prototype decision support system for breast cancer, which is now being developed as a web-based system funded by BioPattern in partnership with Gap Infomedia a leading web development company.

The group were also successful in securing commercialisation funding of £15,000 to allow Ian to work in collaboration with Johnny Rook Ltd, a company who build rich and interactive embeddable web-server analytics and executive dashboards to the web, for OLAP on the Microsoft Platform. Our responsibilities are to deliver an automatic time-series forecasting tool using neural network and OSRE technologies for use by suppliers to a multi-national retailer. In addition, in December 2006 Terence was awarded an Enterprise Fellowship £5,000 in order that he can work on the Johnny Rook Ltd OLAP and Neural Network collaboration.

Prof Paulo Lisboa, Dr Corneliu Arsene and Dr Hane Aung have developed and applied an extended variant of the PLANN-ARD prognosis estimator which can model multiple events (or competing risks) of interest simultaneously. This model known as PLANN-ARD-CR was developed in collaboration with Prof Elia Biganzoli’s group in Milan. A conference paper was accepted for MEDSIP 2006 in Glasgow that demontrated PLANN-ARD-CR applied to patients treated for Uveal Melanoma.

Prof Lisboa and Dr Hane Aung have further investigated compensation and weighting effects during the training of the single risk PLANN-ARD model which led to a further MEDSIP 2006 publication.

In collaboration with Prof Sabine Van Huffel, Dr Ben Van Calster and Prof Dirk Timmerman from Katholieke Universiteit Leuven in Belgium, further application of the OSRE rule extraction software was implemented to gain insight into classification models for Ovarian Tumour Malignancy. The data used is known as the IOTA phase 1 dataset collected by nine different centres form the UK, Italy, Belgium, Sweden and France collectively known as the International Ovarian Tumour Analysis group. A paper has been accepted to appear the proceedings of the International Symposium for Neural Networks in 2007 which will appear in Lecture notes in Computer Science.

Collaboration has also started for the first time with Dr Geert Postma’s group from Radboud University, Nijmegen in the Netherlands. This work involves rule extraction into data extrapolated from MRI and MRSI imaging data on Brain tumour patients.

Further prognosis modelling of breast cancer patients who are specifically HER2 positive is also being undertaken by Dr Hane Aung, Prof Paulo Lisboa and Dr Ian Jarman as a joint project with the THEMIS and Leon Berard cancer research centres in France.

Page last modified by Warren Anacoura on 22 September 2008.
LJMU Logo banner imageLJMU Logo banner image (print)
LJMU banner image
LJMU Dream, Plan Achieve - Page ID:95164