In a preliminary study involving 100 women, the system correctly predicted in almost nine out of ten patients whether the disease would spread to other parts of their body and whether they would survive for five years without a recurrence of cancer.
Dr Gajanan Sherbet and Dr Raouf Naguib, who developed the system at Newcastle University, England, say they are hopeful that it could one day save lives by helping specialists decide at an early stage which patients should have intensive treatment.
The research, funded by the charity Cancer Research UK and reported in this week's New Scientist magazine (publication date Saturday 27 July 2002) also suggests that some of the statistical methods currently used for predicting the outcome of breast cancer may be unreliable.
Working in Newcastle University's Department of Electrical and Electronic Engineering in collaboration with Coventry University's School of Mathematical and Information Science, the two scientists assembled a powerful combination of neural networks, in which the computer's circuits are modelled on the nerves and neurones of the brain, and fuzzy logic, which was originally developed as a means of modelling the uncertainty of natural language.
Once assembled, the system was 'trained' to analyse images of cells from tissue samples, taken from breast cancer patients, for patterns of abnormality which could be used to predict the outcome of the disease.
Breast cancer itself is relatively easy to treat but there is a danger that it might spread via the lymph glands to other parts of the body. Biopsies of lymph glands are routine but the cancer does not always show up in the early stages.
The technique developed by Dr Sherbet and Dr Naguib is an extension of an existing technique called image cytometry - computer analysis of tumour tissue taken from patients and photographed under a microscope.
The two scientists programmed their computer to measure four 'indicators' of how aggressive the cancer might be: The proportion of cells that had extra DNA, the pattern of DNA levels in the whole sample, the number of cells that were dividing and the shape of the cell nuclei.
The information was fed into the neural network, which is especially good at spotting patterns, and then subjected to fuzzy logic, which 'weights' the data to make it fit the patterns as closely as possible.
Dr Sherbet and Dr Naguib trained, or calibrated, the system using tissue samples from 50 breast cancer sufferers and data about the outcome of the cases, such as recurrence of the cancer and the five year survival rate.
Data from a further 50 cases were then fed into the computer, which was asked to predict which of the women would develop tumours in their lymph glands. It did so with 88 per cent accuracy and achieved a similar figure when asked to predict which women would still be alive after five years.
Dr Sherbet and Dr Naguib, together with collaborating scientists in Milan, Italy, claim that their technique of combining fuzzy logic with neural networks is more sophisticated than anything currently available.
'We believe that this technique has produced more reliable prognostic factor models than those obtained using either the statistical or artificial neural networks-based methods,' states the research team in the Journal of Anticancer Research, where it reports its findings.
'In addition, the results suggest that the existing statistical methods may not be reliable as far as both identification of significant prognostic factors and prediction of disease development in the case of individual patients are concerned.'
Dr Sherbet, who has a medical background, commented: 'The results are promising but they would need to be confirmed by full-scale trials before the technique could be introduced into hospitals.'
Dr Sherbet is also a Professor at the Institute of Molecular Medicine in California, USA, but conducted the research at Newcastle with Dr Naguib, who was recently appointed as a Professor of Biomedical Computing in the School of Mathematical and Information Sciences at Coventry University, England.
- Fuzzy logic is based on conventional (Boolean) logic that has been extended to handle the concept of partial truth - truth values between 'completely true' and 'completely false'. It was originally introduced in the 1960s as a means to model the uncertainty of natural language.
- Neural networks are modelled on the brain and are therefore characterised by the ability to learn from experience. Scientists therefore talk about 'training' the neural network, whereas a conventional circuit would be 'calibrated'.
Journal
Anticancer Research