Prostate Cancer Biopsy Recommendation through Use of Machine Learning Classification Techniques

摘要

This study proposes the investigation and application of machine learning techniques in order to aid prostate cancer diagnosis through classification in order to either recommend or spare patients from biopsy, an essential procedure for confirmation of diagnosis. Pre-treatment variables collected from patients of the Academic Hospital of State University of Londrina, Brazil (HU-UEL) from 2005 to 2010 include age, PSA (prostate specific antigen) marker, DRE (digital rectum examination), free/total PSA and PSA density value. Models have been generated using logistic regression, two artificial neural networks (MultiLayer-Perceptron, MLPClassifier) and two decision tree algorithms (ADTree, PART). Obtained accuracy indicators for models were 69.4%, 70.5%, 71.14%, 71.8% and 71.48 % respectively.