Survival Prognosis in Ovarian Cancer

Abstract

In collaboration with Birmingham’s City Hospital we want to attempt a study of likely factors which can provide the medical professionals with better prognosis of ovarian cancer. Current data analysis methods have concentrated upon linear factor analysis to try and identify the most useful prognostic indicators. This project researches and develops advanced pattern processing techniques to try and estimate the likely survival probabilities. In the first part of the project a certain number of methods have been researched to cope with missing data. Then the neural networks approach was introduced: it deals with both regression problems such as estimating how many months a patient is going to live, and classification problems such as finding the probability a patient will die before a given number of months. In the third part confidence in the results obtained is discussed through the analysis of Bayesian error bars and the plotting of ROC curves. The conclusions derived from these analysis are discussed at the end of the thesis.