SBIR-STTR Award

Neural networks (NN) have become established as powerful tools for complex pattern recognition problems. One application which appears well suited to NN methods is the identification of prognostic groups, to be used for treatment planning. For many cancer, studies of cancer cell biology have added many factors of potential prognostic value, but the way in which these interact with known factors is generally not well studied. The potential of NNs to model these data in a non-linear fashion has only begun to be explored. NNs are not part of standard statistical packages, making them relatively inaccessible to many statisticians. More importantly, current NN methods cannot accommodate censored outcome variables. This proposal is for development of algorithms for censored-data NNs, implementation of these within a comprehensive statistical package, and evaluation of alternative approaches. The aim is to provide statisticians involved with clinical decision making with more ready access to NN technology, and with the means to analyze survival-type data. The value of NNs in this field cannot be addressed by any single investigator, but by providing the software that is needed, and some guidelines for its use, we anticipate that research in this field will be stimulated.

Neural networks (NN) have become established as powerful tools for complex pattern recognition problems. One application which appears well suited to NN methods is the identification of prognostic groups, to be used for treatment planning. For many cancer, studies of cancer cell biology have added many factors of potential prognostic value, but the way in which these interact with known factors is generally not well studied. The potential of NNs to model these data in a non-linear fashion has only begun to be explored. NNs are not part of standard statistical packages, making them relatively inaccessible to many statisticians. More importantly, current NN methods cannot accommodate censored outcome variables. This proposal is for development of algorithms for censored-data NNs, implementation of these within a comprehensive statistical package, and evaluation of alternative approaches. The aim is to provide statisticians involved with clinical decision making with more ready access to NN technology, and with the means to analyze survival-type data. The value of NNs in this field cannot be addressed by any single investigator, but by providing the software that is needed, and some guidelines for its use, we anticipate that research in this field will be stimulated.