Neural Network Prediction of Prostate Cancer Progression

点击翻译按钮获取中文摘要

DESCRIPTION (provided by applicant): The broad, long-term objective of this application is to treat prostate cancer effectively through accurate diagnosis and accurate assessment of tumor progression. The hypothesis to be tested in this project is that artificial neural networks (ANNs) can accurately predict prostate cancer progression. The significance and health-relatedness of this research is that accurate prediction of prostate cancer progression is important to identify patients with organ-confined prostate cancer for whom surgery is highly effective, and patients with more advanced prostate cancer for whom surgery is less effective but imposes unnecessary risks of complications who are more appropriate to receive radiation, hormonal, and other therapies. Previous investigation of ANNs often rely on highly-selected ANNs that do not prove or disprove the effectiveness of ANNs. This application will determine, ultimately, whether ANN is more accurate than multivariate linear regression in the prediction of prostate cancer progression. Appropriate statistical models are important to combine clinically the results of an array of biomarkers and cancer predictors. The specific aims are: (1) To develop an ANN-based method for the prediction of pathologic stage and to compare with the Partin nomogram - a clinically accepted multivariate linear regression-based method. (2) To develop a novel method that will add 95% confidence intervals to the ANN prediction of prostate cancer progression. (3) To develop an ANN-based model for the prediction of pathologic stage based on preoperative serial PSA measurements. The research design is to develop ANN-based predictive models and compare them to a clinically accepted standard and previously published results that were considered promising but had produced limited clinical use. The methods to be used include collection of a clinical database, analysis of artificial neural network and multivariate linear regression models, receiver operating characteristic (ROC) analysis, statistical estimation, and computation of confidence intervals.