ABSTRACT
This chapter describes how a neural network was hand-crafted, with considerable fine–tuning, by a computer science postgraduate as part of an individual MSc project. The neural network model used was a standard three layer feed-forward network, trained with the back-propagation algorithm. A criticism of neural network analysis is its black box nature, reflecting the lack of symbolic and hence human-interpretable output. Neural networks lack symbolic output in a form meaningful to domain experts. Machine learning, and neural networks in particular, are increasingly being used in mainstream clinical practice. Poor statistics relating to the incidence and mortality of oral cancer have received considerable attention within the dental profession and there have been repeated calls for screening and prevention of oral cancer. In 1991 a UK working group on screening for oral cancer and precancer was established, with the remit of examining the current literature and advising on the feasibility of screening for oral cancer in the UK.
