Prostate cancer is one of the most common cancers among men over the age 50. Staging serves as a road map for the treatment selection and helps in differentiating organ-confined cancers from the non-organ confined ones (those which have already spread to the outside of prostate-seminal vesicles, lymph nodes and bones). Outcome of treatment varies significantly between these different kinds of cancers. Only organ confined cancers are amenable to curative intent of surgery and radiation therapy. The main objective of this proposal is to explore the neural network technique for the prediction of certain features indicative on non-organ confined prostate cancer on the basis of the results of certain diagnostic tests administered to patients suffering from prostate cancer. The approach is going to be Bayesian, and the goal is to provide the posterior (or predictive) probabilities of the presence of these features in the patients based on certain inputs. The doctors can then make decisions on the basis of these probabilities, and in particular, in marginal cases (for example, when these posterior probabilities are in the neighborhood of 50 percent) go for further diagnostic tests rather than making an immediate decision of whether or not to suggest surgical intervention. Within the Bayesian framework, several methods, both parametric and nonparametric, will be compared. Also, the Bayesian procedures will be compared against some classical frequentist procedures. It may be added here that the Bayesian neural network methods to be proposed are versatile enough, and can be adapted for other nonlinear modeling.
| Ghosh, Malay; Papageorgiou, Georgios; Forrester, Janet (2009) Multivariate Limited Translation Hierarchical Bayes Estimators. J Multivar Anal 100:1398-1411 |
