Method to aid the diagnosis of prostate cancer using machine learning and clinical data

Prostate cancer is the most common type of cancer among men and the one that causes the most deaths in the world. To start the diagnosis of prostate cancer, basically are used digital rectal examination (DRE) and prostate-specific antigen (PSA) levels. Currently, the biopsy is the only procedure able to confirm cancer, it has a high financial cost, and it is a very invasive procedure. In this research, a new method is suggested to aid in the screening of patients at risk of prostate cancer. The method was developed based on clinical variables (age, race, diabetes mellitus (DM), alcoholism, smoking, systemic arterial hypertension (SAH), DRE, and total PSA) obtained from the patient’s medical records. The method was tested using the algorithms of machine learning: Support Vector Machine (SVM), Naive Bayes (NB), K-Nearest Neighbor (KNN), Decision Trees (DT), and Artificial Neural Networks (ANN), which predicted the samples between the presence or absence of prostate cancer. The method evaluation was made by performance metrics: accuracy, specificity, sensitivity, and AUROC (area under the receiver operating characteristic). The best performance found was through the Linear SVM model, resulting in an accuracy of 86.8%, sensitivity of 88.2%, specificity of 85.3%, and AUROC of 0.90.