Abstract Code: IUC26651-85
AI-assisted Biparametric vs Multiparametric MRI for Clinically Significant Prostate Cancer
- Tiwari 1, K. Bhagat 2, U. Khan 3, J.K. Tan 4, D. Sethi 5, A. Maniam 6, G. Banna 7, A. Khan 8, A. Ghose 1
(1) Cancer AI Validation Lab, OncoFlow – United Kingdom, (2) Royal Surrey Cancer Centre, Royal Surrey NHS Foundation Trust – United Kingdom, (3) Early Cancer Institute, University of Cambridge – United Kingdom, (4) School of Medical Sciences, University of Manchester – United Kingdom, (5) University of Buckingham Medical School – United Kingdom, (6) Department of Oncology, Isle of Wight NHS Trust – United Kingdom, (7) Department of Oncology, Portsmouth Hospitals University NHS Trust – United Kingdom, (8) Department of Radiology, Leeds Teaching Hospitals NHS Trust – United Kingdom
Background
Multiparametric MRI (mpMRI) remains the pre-biopsy reference standard for detecting clinically significant prostate cancer (csPCa, ISUP Grade ≥2). Biparametric MRI (bpMRI) provides a faster, lower-cost alternative that may reduce barriers to pre-biopsy imaging. Artificial intelligence (AI) may improve diagnostic performance in both modalities. We systematically reviewed AI models for csPCa detection using bpMRI and mpMRI.
Methods
A systematic review (PROSPERO ID: CRD420251037432) searched MEDLINE, PMC, EMBASE, SCOPUS and COCHRANE. Of 6,389 records, 202 underwent full-text review; with six meeting inclusion criteria. Studies were evaluated using TRIPOD-SRMA and TRIPOD-AI.
Results
Six studies compared AI-assisted csPCa diagnosis using bpMRI (PMID: 40259798, 36222324, 38876123) and mpMRI (PMID: 40016318, 37345961, 33671533). Considerable methodological heterogeneity was observed, particularly in data preparation, sample size estimation, fairness assessment, and external validation.
Table 1. Comparative diagnostic performance of AI-assisted bpMRI and mpMRI for csPCa
Performance Metric | AI-assisted mpMRI (pooled) | AI-assisted bpMRI (pooled) | Human-only mpMRI (PROMIS trial) | AI-assisted bpMRI (PI-CAI benchmark) |
Sensitivity | 0.88 (0.75-0.98) | 0.65 (0.39, 0.84) | 0.93 (0.88, 0.96) | 0.97 (0.95, 0.98) |
Specificity | 0.68 (0.51-0.80) | 0.93 (0.69, 0.99) | 0.41 (0.36, 0.46) | 0.50 (0.42, 0.58) |
AUC | 0.84 (0.69-0.95) | 0.88 (0.80, 0.93) | Not Reported | 0.92 (0.89, 0.94) |
AI-assisted bpMRI demonstrated higher specificity than mpMRI, although with lower sensitivity. This indicates a potential trade-off between detecting all cases of clinically significant prostate cancer and minimising false-positive findings, which may be influenced by differences in model training approaches and the simplified imaging protocol used in bpMRI.
Conclusion
AI-augmented MRI is emerging as a radiologist-assistive tool with potential to reshape pre-biopsy prostate cancer imaging. AI-assisted bpMRI may offer comparable diagnostic accuracy to mpMRI while reducing acquisition time and cost, supporting implementation in high-throughput or resource-limited settings. Conversely, AI-assisted mpMRI may improve precision in diagnostic imaging at tertiary centres. Future work should prioritise external validation, multi-centre calibration, and assessment of real-world clinical utility to ensure safe, equitable integration of AI into prostate cancer diagnostic workflows.