Artificial intelligence-enabled MRI surveillance of prostate cancer: integrating PRECISE v2, PI-QUAL, and longitudinal biomarkers-a narrative review.
Authors
Affiliations (10)
Affiliations (10)
- Department of Pathology, Memorial Sloan Kettering Cancer Center, New York, USA. [email protected].
- Graduate School of Medical and Dental Sciences, Institute of Science Tokyo, Tokyo, Japan. [email protected].
- Division of Surgical and Interventional Science, University College London, London, UK.
- Department of Urology, University College London Hospitals Trust, London, UK.
- Department of Radiology, Giresun University, Giresun, Turkey.
- Department of Radiology, University Medical Center Groningen, Groningen, Netherlands.
- Graduate School of Medical and Dental Sciences, Institute of Science Tokyo, Tokyo, Japan. [email protected].
- Department of Radiology, University of California, San Diego, San Diego, USA. [email protected].
- Imaging Institute, Abdominal Imaging Section and Department of Nuclear Medicine, Diagnostics Institute, Cleveland Clinic, Cleveland, USA.
- Departments of Urology and Colorectal Surgery, Cleveland Clinic, Cleveland, USA.
Abstract
Active surveillance (AS) is widely used for men with low-risk and selected favorable intermediate-risk prostate cancer, but pathways remain heterogeneous and often rely on serial biopsy despite sampling limitations and morbidity. Multiparametric MRI (mpMRI) improves enrollment and longitudinal monitoring, yet reproducibility depends on image quality, reader expertise, measurement consistency, biopsy quality, and clear definitions of progression. The PRECISE recommendations were developed to standardize serial MRI reporting, and PRECISE version 2 (PRECISE v2) introduces key refinements, including minimum quality thresholds (PRECISE-X), comparison with both baseline and prior scans, and subdivision of stable examinations into visible and non-visible disease (3-V and 3-NonV). In parallel, artificial intelligence (AI) for prostate MRI is advancing in quality control, detection, segmentation, and longitudinal analysis. Near-term opportunities in AS include automated quality checks, reduction of measurement variability through segmentation and lesion tracking, and integration of imaging change with clinical and biomarker trajectories such as PSA density and kinetics. However, key barriers include dataset shift, sample-size imbalance, imperfect progression labels, limited external validation, radiomic feature instability, calibration, missing data, uncertainty quantification, and the need for transparent human-AI interaction. This narrative review summarizes PRECISE v2, its relationship with PI-RADS and PI-QUAL, and emerging strategies to combine PRECISE-aligned imaging change with longitudinal clinical and biomarker information for risk-adapted decision support. Serial mpMRI can support AS when image quality and serial reporting are standardized; however, MRI should complement rather than replace biopsy, and AI-assisted pathways require AS-specific external validation, calibration, harmonization, and prospective evaluation before they can be used to modify biopsy timing.