Generative AI Low-Dose DSA Cuts Radiation In Randomized Trial

Key Takeaways:

• A randomized multicenter trial reported lower intraoperative radiation exposure with GenDSA-V2 during digital subtraction angiography-guided procedures.
• Mean operation time was reported as noninferior, and intraoperative complication rates were reported as similar between groups.
• Source data, supplementary materials, and model-construction code were reported as available through the listed access pathways.

They reported lower radiation exposure with GenDSA-V2 than with standard clinical protocols during DSA-guided procedures. GenDSA-V2 was prospectively validated in enrolled patients undergoing these procedures. The enrolled groups covered three indications—suspected cerebral aneurysm, lung cancer, and advanced liver cancer—under surgical criteria. The study focused on radiation reduction during real-world procedural imaging.

The multicenter randomized controlled trial enrolled 1,068 patients: 533 assigned to GenDSA-V2 and 535 assigned to standard protocols. Patients, surgeons, and investigators were blinded to group assignment; technicians knew assignments but did not collect or analyze study data. Radiation dose was the primary outcome, with procedural efficiency, operation time, and intraoperative complications assessed as secondary outcomes. Enrollment spanned all three disease groups under the surgical criteria used for validation. The protocol paired masked assignment with prespecified radiation and procedure-related outcomes.

Reported radiation metrics were lower in the GenDSA-V2 group than in the standard-protocol group. Air kerma was 151.3 ± 125.1 mGy with GenDSA-V2 versus 457.4 ± 407.4 mGy with standard protocols. Dose-area product was 4009.7 ± 2767.9 μGy m2 versus 12531.6 ± 9145.9 μGy m2, respectively. Both comparisons met superiority testing (air kerma mean difference −306.1 mGy, 95% CI −342.3 to −269.9; dose-area product mean difference −8521.9 μGy·m², 95% CI −9333.1 to −7710.7; P<0.001 for both).

The authors also reported secondary findings on procedural efficiency and safety. Mean operation time was 34.8 ± 11.8 minutes with GenDSA-V2 and 33.1 ± 10.8 minutes with standard protocols. This was reported as noninferior, with a mean difference of 1.7 minutes (95% CI 0.3 to 3.1) and P<0.001 for noninferiority. Intraoperative complication rates were similar (7.5% vs 8.1%; mean difference −0.6%, 95% CI −3.8% to 2.6%), with noninferiority reported (P<0.001). In the reported analysis, procedural duration and intraoperative complications did not materially separate the two groups.