Background: Comprehensive collection and analysis of stroke imaging data are essential for refined clinical decision-making and research in patients with ischemic stroke. However, the voluminous and heterogeneous nature of imaging data has historically posed significant challenges to systematic prospective collection in multicenter registries. Method: Building upon the Clinical Research Collaboration for Stroke in Korea (CRCS-K), a nationwide prospective multicenter acute stroke registry, all brain imaging studies performed for clinical purposes during hospitalization of consecutive patients with acute ischemic stroke were systematically collected. At each participating institution, imaging data underwent primary de-identification before secure transfer to a central imaging laboratory at the coordinating center, either via online transmission or offline delivery. The central imaging laboratory subsequently performed pseudonymization, quality assurance, modality and sequence classification, and sequence-level tagging. Quantification of stroke imaging was conducted using automated artificial intelligence (AI) pipelines, and processed datasets and images were then distributed to participating investigators. Results: Between the second half of 2022 and May 2025, stroke imaging was prospectively collected from 20,398 (92.9%) out of 21,961 consecutive ischemic stroke patients admitted to 18 participating hospitals (Figure 1). Among patients with available imaging, 60% were male, the mean age was 69 ± 14 years, and the median arrival to first imaging time was 0.48 hours (IQR, 0.13–1.39). Magnetic resonance imaging was the initial modality in 16% of cases (Table 1). Based on detailed sequence classification, the repository comprises 50,561 computed tomography sequences and 171,836 magnetic resonance sequences. Leveraging this dataset, the central imaging laboratory provides participating researchers with diverse quantified imaging information through automated AI-supported pipelines (Figure 2). Conclusion: The CRCS-K imaging repository demonstrates the feasibility of prospective, large-scale collection of stroke imaging in a multicenter setting, achieving a 92.9% capture rate with integrated AI-supported analytic pipelines. This infrastructure constitutes a valuable resource for standardized imaging analysis and has the potential to facilitate future research in nuanced treatment and outcome prediction.
Kim et al. (Thu,) studied this question.