Enhanced PSA Density Prediction Accuracy When Based on Machine Learning

PurposeThe prostate-specific antigen (PSA) density (PSAD) in prostate cancer (PCa) detection has limited applicability and is probably caused by moderate accuracy. The purpose of this study was to create a machine learning (ML) PSAD model that incorporates PSAD predictors for forecasting clinically significant (cs) prostate cancer (PCa) probability and compare its performance to that of the traditional PSAD.MethodsPSA and prostate volume (PV) were retrieved from the 725 patients that were subjected to prostate biopsy. After resampling and splitting data, we used the training set to create seven ML algorithms. We chose the RF model that was the most accurate. The area under the curve (AUC) accuracy, precision, sensitivity, and specificity of PSAD and RF PSAD diagnostic performance were compared. Additionally, the ML model's explainability and its website placement were performed.ResultscsPCa was found in 140 males (19.3%). The proposed novel model exhibited much higher evolution metrics than PSAD. AUC for the PSAD and RF PSAD were 0.757 and 0.942, respectively. The reliability diagram indicates that the RF model fits the data well. For the RF model, the decision curve analysis revealed a net benefit of more than 5%, and 40% subjects could avoid unnecessary biopsy. PV was the more important determinant for csPCa. PSA and PV had non-monotonic relationships and a lot of turbulence.ConclusionThe RF PSAD model demonstrated strong discrimination and clinical value, which could aid urologists in determining whether a prostate biopsy is required.