MRI-ultrasound fusion AI model shows higher specificity than radiologist MRI interpretation for prostate lesions

Study · BJU Int · 2026-10-21. Number 2 in the UrologyWatch Top 5 for October 21, 2026.

In a radical prostatectomy validation cohort (110 patients), the MRUSuNet deep learning model, which combines MRI and TRUS, showed significantly higher specificity (0.88 vs 0.78) and lesion-level Dice overlap (0.32 vs 0.27) than radiologist MRI interpretation, with no significant difference in sensitivity. MRUSuNet also outperformed MRI-only and TRUS-only models in sensitivity and Dice score across test cohorts.

Why it matters: Suggests AI-based MRI-ultrasound fusion could improve lesion localization accuracy versus radiologist interpretation, though validation beyond study cohorts and prospective testing are needed before clinical use.

Limitation: Development and validation used specific institutional cohorts; real-world generalizability and prospective biopsy outcomes not yet established.

Source: BJU Int · PubMed 42901865 · DOI 10.1111/bju.70491

The full October 21, 2026 edition · Today's urology news