MRI plus ultrasound AI localised prostate cancer more specifically than radiologists

Design

Deep-learning development and validation study

Population

3,000 fusion-biopsy patients for development; 110 prostatectomy patients for validation

Primary outcome

Sensitivity, specificity and lesion Dice score for significant cancer

Effect

Sensitivity 0.80 vs 0.73 (MRI-only); specificity 0.88 vs 0.78 for radiologists

In MRI-ultrasound fusion biopsy, misregistration between the two images can make the needle miss a significant cancer. A BJU International study trained a deep-learning model that takes both MRI and transrectal ultrasound as input to outline significant cancer directly on the ultrasound image, using fusion-biopsy cohorts of 3,000 men, with 1,700 held-out cases and an independent prostatectomy cohort of 110.

Across test cohorts, the combined model reached sensitivity 0.80 against 0.73 for an MRI-only model and 0.49 for ultrasound-only, with better lesion overlap. In the prostatectomy cohort it matched radiologists' sensitivity (0.79) and had higher specificity (0.88 against 0.78).

The comparison with radiologists rests on 110 men. It is a development study, but it addresses a real weakness of fusion biopsy and is worth watching as these tools move towards the biopsy suite.

• A combined MRI-ultrasound AI model detected significant cancer more sensitively than MRI-only or ultrasound-only models.
• Against radiologists, it had higher specificity with similar sensitivity on prostatectomy validation.
• The tool is investigational and not available for clinical biopsy targeting.
• Continue to sample around MRI targets to allow for registration error during fusion biopsy.

Why it matters

Registration error between MRI and ultrasound is a hidden source of missed cancers that AI may reduce.

Don't overread it

The radiologist comparison involved 110 men and has not been tested prospectively in live biopsies.

Read the rest in the app

You have read your two free briefings this month. The app carries all 27 specialties, every morning, free — and this finding is waiting in it.

Get Daily Doctor — free

Scan to keep reading on your phone. No account needed to start.