Prediction of Prostate Cancer Risk Stratifications Based on A Non-Linear Transformation Stacking Learning Strategy

Abstract

Abstract Prostate cancer (PCa) is an epithelial malignancy that occurs in the prostate gland and is generally classified into three risk categories: low, intermediate, and high risk. The most important diagnostic indicator for PCa is the measurement of serum prostate-specific antigen (PSA) values, but this method can produce false positives leading to unnecessary biopsies, increasing the likelihood of invasive injuries. Therefore, it is imperative to develop an efficient and accurate method to predict PCa risk stratifications. Most current studies on predictions of PCa risk stratification based on clinical data generally perform only a dichotomy of low to intermediate and high risk. This paper proposed a novel machine learning (ML) approach based on a Stacking learning strategy to predict tripartite risk stratifications of PCa. Clinical records with features selected by Lasso were learned by five ML classifiers. Outputs of five classifiers were transformed by various nonlinear transformers (NT) and then, concatenated with the Lasso-selected features to obtain a set of new features. A Stacking learning strategy integrating different ML classifiers was developed based on these new features. Our proposed approach achieved superior performance with an accuracy (ACC) of 0.83 and an Area Under the Receiver Operating Characteristic curve (AUC) value of 0.88 in a dataset of 197 PCa patients with 42 clinical characteristics. This study will better assist clinicians in rapidly assessing PCa risk stratifications while reducing patient burden through AI-related technologies in auxiliary diagnosis of PCa.