DOI: 10.1161/jaha.126.050940 ISSN: 2047-9980

CRCS‐K Imaging Repository: Architecture, Data Characteristics, and a Proof‐of‐Concept Analysis of Pretreatment Imaging Workflow

Beom Joon Kim, Wi‐Sun Ryu, Myung Jae Lee, Kyusik Kang, Jae Guk Kim, Soo Joo Lee, Jae‐Kwan Cha, Tai Hwan Park, Jeong‐Yoon Lee, Kyungbok Lee, Doo Hyuk Kwon, Jun Lee, Hong‐Kyun Park, Yong‐Jin Cho, Keun‐Sik Hong, Minwoo Lee, Mi Sun Oh, Kyung‐Ho Yu, Dong‐Seok Gwak, Dong‐Eog Kim, Hyunsoo Kim, Joon‐Tae Kim, Joong‐Goo Kim, Jay Chol Choi, Wook‐Joo Kim, Young Cheol Weon, Jee Hyun Kwon, Kyu Sun Yum, Dong‐Ick Shin, Jeong‐Ho Hong, Sung‐Il Sohn, Sang‐Hwa Lee, Chulho Kim, Hae‐Bong Jeong, Kwang‐Yeol Park, Chi Kyung Kim, Jihoon Kang, Jun Yup Kim, Do‐Yeon Kim, Jonguk Kim, Nakhoon Kim, Leonard Sunwoo, Sung Hyun Baik, Bijoy K. Menon, Longting Lin, Mark Parsons, Hee‐Joon Bae
Show PDF Cite

Background

Stroke registries have advanced cerebrovascular research but usually reduce neuroimaging to categorical variables, losing multidimensional information. We describe the CRCS‐K (Clinical Research Collaboration for Stroke in Korea) Imaging Repository, a multicenter platform integrating stroke imaging, artificial intelligence‐based quantification, and clinical and outcome data through a dedicated research platform, Artificial Intelligence Powered Stroke Clinical‐Image Archive Network (AISCAN).

Methods

Building upon the nationwide CRCS‐K registry, the Imaging Repository collected all computed tomography, magnetic resonance, and angiographic studies obtained during index hospitalization from consecutive patients with acute ischemic stroke at 18 stroke centers. Images underwent centralized deidentification, quality verification, sequence classification, and artificial intelligence‐based quantification. As a proof‐of‐concept application, we examined associations of pretreatment imaging modality with treatment workflow and functional outcomes after intravenous thrombolysis or endovascular treatment.

Results

From June 2022 through May 2025, 225 159 imaging sequences were collected from 20 792 patients. AI modules generated standardized numeric features including ischemic lesion volumes, perfusion parameters, white matter hyperintensity burden, and cerebral microbleed counts. Magnetic resonance‐first workflows varied substantially across centers, from 1.0% to 56.7%. Greater pretreatment imaging sequence burden was associated with longer door‐to‐treatment times after intravenous thrombolysis and endovascular treatment. In overlap‐weighted analyses, magnetic resonance‐based versus computed tomography‐based imaging was associated with directionally lower, but not statistically significant, odds of a favorable 3‐month outcome after intravenous thrombolysis (odds ratio [OR], 0.90 [95% CI, 0.70–1.18]) and endovascular treatment (OR, 0.89 [95% CI, 0.65–1.21]).

Conclusions

Prospective, sequence‐level retention of all stroke neuroimaging is feasible at network scale and can be integrated with artificial intelligence‐derived features and clinical outcomes. This resource enables real‐world investigation of imaging workflows and outcomes.

More from our Archive