- Design
- Retrospective reader study of a deep-learning virtual DCE model
- Population
- 592 3 T breast MRI examinations; 2 board-certified radiologists
- Primary outcome
- BI-RADS diagnostic accuracy, sensitivity and specificity
- Effect
- Accuracy 76% vs 77%; sensitivity 92% vs 96%; specificity 65% vs 64% (all p = 1.0)
Gadolinium-enhanced dynamic imaging is the core of breast MRI. A retrospective European Radiology study tested a deep-learning model that generates virtual dynamic contrast-enhanced images from unenhanced T1-weighted, T2-weighted and diffusion-weighted sequences, using 592 3 T examinations split into training, validation and test sets.
Two board-certified radiologists read virtual and standard examinations with a BI-RADS-based protocol. Mean diagnostic accuracy was 76% for virtual against 77% for standard imaging; sensitivity was 92% against 96% and specificity 65% against 64%. None of these differences reached statistical significance. Lesion size differed slightly and inconsistently between test sets (−2.2 mm and 0.5 mm), background enhancement and artefact ratings were similar, and one reader was less confident with virtual images.
The authors' own conclusion is cautious, and rightly so. A four-point sensitivity gap is not statistically significant in a study of this size, but in screening or staging it would mean missed cancers. This is not yet a replacement for contrast, though it may in time suit selected patients who cannot receive gadolinium.
- Virtual contrast breast MRI is investigational; continue to use gadolinium-enhanced protocols for diagnosis and screening.
- Sensitivity was 92% with virtual images against 96% with standard contrast in this study.
- Lesion size on virtual images differed by up to about 2 mm, enough to matter for staging.
- Patients asking about contrast-free breast MRI can be told the technique is promising but not yet validated.
Why it matters
Even small sensitivity losses in breast MRI translate into missed cancers in high-risk women.
Don't overread it
Lack of statistical significance here reflects a small reader study, not equivalence between virtual and standard contrast.
The statistics, in plain English
A non-significant difference in a study with two readers does not show the methods are the same. A four-point fall in sensitivity means about four more missed for every hundred cancers present.
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.
Scan to keep reading on your phone. No account needed to start.
Tomorrow morning, before your first patient
One edition a day for radiology, written by the desk, every claim tied to its paper. Six minutes.
Get the app — free