Abstract for: Early Diagnosis of Prostate Cancer by Dynamic Modeling and Data Science Tools

Studies have developed alternative tools to detect prostate cancer in its early stages; however, their scope remains limited because of the strong assumptions they hold, resulting from the limitations in the medical literature. In our project, we study tissue-level dynamics of prostate, and we model the potential tumor presence and dynamics using two methodologies: system dynamics and data science. Objective of the study is to come up with an improved diagnosis method supported by two models. We build the dynamic model using stock-flow modeling and simulation to observe the time-dependent dynamics in the prostate. To fill the missing parts of data obtained from the literature, we make use of the dynamic model to produce synthetic data to be used as an input in the machine learning models. Using Python, we build nine different classification models and XGBoost Classifier performs the best among others with an accuracy value of 81.75 and recall value of 87.71. Both models are validated using available real-world data on prostate cancer. Combined outputs from two models provide added information on tumoral status and processes in a given individual. This study can be eventually useful to improve the medical screening procedures towards early diagnosis of prostate cancer.