Objective: To assess large language models’ (LLMs) mastery of contemporary consensus documents in guideline-based neurointerventional decision-making domains and to compare their performance with radiologists of different experience levels. Methods: In this cross-sectional study, 100 single-best-answer, 5-option MCQs were derived from 3 CIRSE guideline sources. A nonparticipating general radiologist authored the items (33 carotid stenting, 33 thrombectomy, 34 acute stroke EVT). One neurointerventional radiologist, 2 general radiologists, and 1 assistant radiologist answered independently, blinded to the key. The same set was posed in January 2026 to Claude 4.5 Opus, ChatGPT 5.2 Instant, and Gemini 3 Pro using default settings and a standardized prompt. Accuracy was compared with McNemar test and Bonferroni correction (P ≤ .002). Results: The highest accuracy was observed with Claude 4.5 Opus (mean [SD], .99 [.10]); ChatGPT 5.2 Instant (.96 [.19]) and Gemini 3 Pro (.95 [.21]) showed comparable performance. Among human participants, the neurointerventional radiologist achieved the highest accuracy (.85 [.35]), while general radiologists scored .67 to .70 and the assistant radiologist achieved .57. Each of the 3 LLMs significantly outperformed all human groups. Moreover, Claude 4.5 Opus also demonstrated significantly higher performance than the neurointerventional radiologist (P = .001). Conclusions: LLMs have reached a competitive level with neurointerventional radiologists in recalling and interpreting neurointerventional guideline knowledge. In particular, Claude 4.5 Opus demonstrated strong performance and appears to be a promising candidate for real-time decision support applications. Multimodal validation and verification studies using real clinical scenario–based designs are warranted.
Background/Objectives: We assessed whether preprocedural serum albumin adds prognostic information to the PCI-derived Mehran score for early contrast-associated acute kidney injury (CA-AKI) after elective carotid artery stenting (CAS). Methods: This single-centre retrospective cohort included 121 consecutive elective patients (January 2020–April 2026). The primary model combined the reconstructed Mehran score with albumin < 3.5 g/dL, a conventional threshold fixed in advance. Because the score uses actual contrast volume, the combined model applies only after CAS. The outcome used KDIGO serum creatinine thresholds (≥0.3 mg/dL or ≥1.5 times baseline) from routine second-day testing. Internal validation used 1000 bootstrap resamples. Results: Early CA-AKI occurred in 32 patients (26.4%). Adjusted odds ratios were 1.35 per Mehran point (95% CI 1.13–1.60) and 11.68 for albumin < 3.5 g/dL (95% CI 4.14–32.97). AUC increased from 0.768 to 0.865 (95% CI 0.794–0.937; likelihood ratio χ2 = 24.58, p < 0.001). Corrected AUC, calibration slope, and Brier score were 0.860, 0.96, and 0.132. In exploratory decision curves (20% threshold), corrected net benefit was 0.168 for the combined model and 0.133 for Mehran alone. Conclusions: Low preprocedural albumin added prognostic information to the PCI-derived score and may help select patients for closer renal surveillance after CAS. External validation and prospective impact studies are needed before clinical use.