Validation of a Machine Learning Model Using Multiparametric MRI for Predicting Clinically Significant Prostate Cancer

What this Trial Is About

Researchers are evaluating a machine learning model designed to predict clinically significant prostate cancer using multiparametric magnetic resonance imaging (RMmp). This observational study aims to validate how well the model identifies true positive and true negative cases by measuring specificity, sensitivity, positive predictive value, and negative predictive value over a five-year period. The study includes data collected both retrospectively and prospectively from patients undergoing prostate imaging and biopsy. The study involves analyzing prostate RMmp images from patients at the IRCCS Bologna University Hospital. Radiologists will use a specialized software tool integrated into an open-source Dicom viewer to independently segment lesions on the ADC sequences and generate a predicted probability that each lesion is clinically significant. The tool collects organized, pseudo-anonymized data for validation. After evaluation of the model's performance and bias, a retraining phase will update the model using a combined dataset from all enrolled participants. Participants will be enrolled by the hospital's radiologist team and must have at least one lesion identified on prostate RMmp images. Researchers will collect imaging and pathology data, including digital scans of prostate tissue when available. The main outcomes measured include the accuracy of the model in predicting significant prostate cancer from enrollment through five years of follow-up. The study excludes patients with prior prostate surgery, hormone therapy, or poor-quality imaging due to artifacts.

Validation of a Machine Learning Model Based on MR for the Prediction of Prostate Cancer