Prostate cancer (PCa) is the most common malignancy in men in both Europe and the United States. Improved treatment and earlier diagnosis have almost halved PCa-specific mortality since the 1990s. However, after the introduction of prostate-specific antigen (PSA), millions of men with clinically insignificant cancer not destined to cause death have received treatment, with no beneficial impact on overall survival. It is well understood that whole gland treatments could be avoided in men with indolent PCa, provided that they are properly selected. 

Since 2020, the European Association of Urology guidelines strongly recommend Magnetic Resonance Imaging (MRI) prior to prostate biopsy to localize cancer and to diagnose extra-prostatic extension. MRI is superior to clinical staging, as it increases detection of PCa and allows a more precise risk classification. Moreover, men with suspicious findings at imaging can benefit from fusion biopsy, merging MRI information with real-time ultrasound (US), providing higher sampling precision and improved diagnostic yield. Unfortunately, MRI of the prostate largely relies on qualitative assessment and suffers from large inter-reader variability, being strongly related to readers’ expertise. Furthermore, qualitative assessment does not allow determination of tumor aggressiveness.

In recent years, efforts have been made to determine if quantitative radiomics signatures could allow better assessment of PCa aggressiveness, using both conventional statistics metrics and higher-order texture features derived from MRI. Moreover, machine learning (ML) methods have been implemented to sift through the large amounts of high-dimensional data provided by radiomics, to optimize accuracy, reproducibility, and throughput.