MRUSuNet AI fusing MRI and ultrasound outperforms single-modality models and improves specificity over radiologists in prostate cancer detection

Prostate cancer AI (MRUSuNet) combining MRI and ultrasound achieved sensitivity 0.80 (vs 0.73 MRI-only, 0.49 US-only) and higher specificity than radiologists (0.88 vs 0.78) on independent radical prostatectomy cohort.

• The model was developed on 3000 patients from two MRI-TRUS fusion biopsy cohorts and externally validated on 110 radical prostatectomy whole-mount specimens.
• MRUSuNet achieved a lesion Dice score of 0.42 overall, significantly higher than MRI-only (0.30, P<0.001) and TRUS-only (0.27, P<0.001).
• Compared with radiologists, sensitivity was similar (0.79 for both), but the AI offered higher specificity (0.88 vs 0.78; P=0.004) and Dice score (0.32 vs 0.27; P=0.048).

Automated summary

RadPigeon summaries are original and for information only. They are not clinical advice.