Laxmi Manasa Gorugantu

Bachelor of Technology (B.Tech), Master of Science (M.S), Ph.D (in progress)

Dakota State University

Doctoral Student in Information Systems at Dakota State University

PLCO (Learn more about this study)

PLCO-873

Dec 13, 2021

Detection of prostate cancer using machine learning techniques: An exploratory study

Prostate cancer (PCa) is one the most frequent and fatal cancers in men. It can be slow-growing and indolent or fast-growing and aggressive. Testing for PCa remains problematic. Evidence is mounting that overdiagnosis and over-treatment can result in adverse side-effects yet have little impact in preventing death from PCa. Consequently, the importance of predictive tools that help physicians in the diagnosis of the condition cannot be understated. There are several prediction models using PSA and other risk factors for detecting clinically significant PCa. However, these models tend to predominantly rely on multivariate logistic regression and tend to be limited in the number of risk factors accounted for in the model. Accordingly, the objective of the research is to investigate the potential of various machine learning techniques using an expanded set of risk factors to improve the sensitivity and specificity of detecting clinically significant PCa. Compared to logistic regression, the machine learning techniques considered could account for the complexity in predicting PCa. Examples of machine such techniques include support vector machines (SVM), decision trees, Bayesian classifiers, and random forest. Risk factors considered include prostate-specific antigen (PSA), digital rectal examination (DRE), as well as age, race/ethnicity, and family history. The proposed model will be evaluated for specificity and sensitivity against state-of-the-art models. By capturing more complex relations between risk factors and incidence of PCa, the resultant predictive model may have the potential for reducing the downstream harms of PSA testing.

1. The objective of the research is to investigate the potential of various machine learning techniques using an expanded set of risk factors to improve the sensitivity and specificity of detecting clinically significant Prostate Cancer.

2. To evaluate the proposed model based on specificity and sensitivity against state-of-the-art models by capturing more complex relations between risk factors and incidence of PCa, the resultant predictive model may have the potential for reducing the downstream harms of PSA testing.