Zhao generative low-dose DSA

Generative AI-based low-dose digital subtraction angiography for intra-operative radiation dose reduction: a randomized controlled trial

Patient / Population Intervention / Exposure Comparison Outcome

In 1,068 patients undergoing DSA-guided procedures for suspected cerebral aneurysm, lung cancer or advanced liver cancer, generative-AI low-dose angiography (GenDSA-V2) reduced intra-operative radiation dose compared with standard clinical DSA protocols.

N
1,068
Design
Multicentre RCT (patients, surgeons and investigators blinded)
Endpoint
Radiation dose (air kerma)
Relevance
3
ResultAir kerma 151.3 ± 125.1 vs 457.4 ± 407.4 mGy (mean difference −306.1 mGy, 95% CI −342.3 to −269.9; P<0.001 for superiority). Operation time 34.8 vs 33.1 min (difference 1.7 min, 95% CI 0.3–3.1; non-inferior); complications 7.5% vs 8.1% (difference −0.6%, 95% CI −3.8% to 2.6%).
Zhao H, et al. Nat Med. 2026;32(1):288-296. 10.1038/s41591-025-04042-6
Discussion & critique

First prospective randomised validation of a low-dose DSA method: a generative model reconstructing diagnostic-quality images from reduced-dose acquisitions cut radiation to patients and operators by roughly two-thirds without lengthening procedures or adding complications. The endpoint is dose rather than a clinical outcome, the model was iterated on a large multicentre image bank from the same group, and technicians were necessarily unblinded.