Evaluation of ChatGPT-5 for Automated ASPECTS Assessment on Non-Contrast CT in Acute Ischemic Stroke

Background/Objectives: This study aimed to evaluate the ability of ChatGPT-5, a multimodal large language model, to perform automated ASPECTS assessment on non-contrast CT (NCCT) in patients with acute ischemic stroke. Methods: This retrospective, single-center study included 199 patients with anterior circulation AIS who underwent baseline NCCT before reperfusion therapy between November 2020 and February 2025. Each NCCT was evaluated by two human readers and by ChatGPT-5 using four representative images (two ganglionic and two supraganglionic). Interobserver agreement was measured with the intraclass correlation coefficient (ICC), and prognostic performance was analyzed using multivariable logistic regression and receiver operating characteristic (ROC) analysis for 3-month functional independence (mRS ≤ 2). Results: ChatGPT-5 demonstrated good-to-excellent agreement with expert consensus (ICC = 0.845; 95% CI, 0.792–0.884; κ = 0.79). ChatGPT-ASPECTS were independently associated with 3-month functional independence (OR = 1.28 per point; p = 0.004), comparable to consensus-ASPECTS (OR = 1.31; p = 0.003). Prognostic discrimination was similar between ChatGPT-5 and consensus scoring (AUC = 0.78 vs. 0.80; p = 0.41). Conclusions: ChatGPT-5 achieved high reliability and strong prognostic validity in automated ASPECTS assessment without task-specific training. These findings highlight the emerging potential of large language models for quantitative image interpretation, though clinical implementation will require multicenter validation and regulatory approval.