Collaborative generalist-specialist AI for safe worklist deprioritization of non-contrast brain computed tomography

Safe deprioritization of non-contrast brain computed tomography (NCCT) demands high sensitivity and negative predictive value (NPV) across common and uncommon findings requiring clinical attention. Here, we present a collaborative safety net integrating BrainCT-CLIP, a generalist foundation model via contrastive language-image pre-training (CLIP), with commercial detectors. Every examination remained in the reporting workflow, deprioritized only when no model returned a positive result.

In a retrospective multicenter external validation of consecutive emergency and trauma cohorts (n = 2001), standalone commercial detectors left 31 of 324 remarkable examinations deprioritized, including 16 target-disease false negatives and 15 outside their indications. In contrast, the collaborative system recovered every remarkable examination at a threshold prespecified in the development cohort, achieving 100% sensitivity and NPV while deprioritizing 18% of examinations (368/2001; 95% CI, 17–20). These findings support generalist-specialist AI for safe worklist deprioritization under radiologist oversight, with high sensitivity for intracranial findings requiring clinical attention.

The authors express their gratitude to Coreline Soft Co., Ltd. and JLK Inc. for providing access to their respective AI algorithms and assisting with their integration into the collaborative workflow. We also thank Ji-Hwan Kim for assistance with data collection and curation, and Su-Jeong Kwon and Eun-Ji Yuk for data collection, curation, and the provisional reference-standard labeling performed under radiologist supervision. The first draft of the manuscript was written by Seungjun Lee, one of the co-authors (Department of Biomedical Engineering, Asan Medical Center).

No paid writing assistance was provided beyond copy editing. During manuscript preparation, the authors used Grok (built by xAI; accessed September–November 2025) for English-language polishing, sentence restructuring, and improving readability in selected sections; the authors reviewed and edited all content generated and take full responsibility for the accuracy, integrity, and originality of the publication. No other authoring, artificial intelligence, machine learning, or large language model tools were used in composing the manuscript.

This research was supported by grants from the Korea Health Technology R&D Project through the Korea Health Industry Development Institute (KHIDI), funded by the Ministry of Health & Welfare, Republic of Korea (HI22C0471, RS-2025-02213531, RS-2022-KH130266); the National Research Foundation of Korea (NRF), funded by the Korean government (MSIT) (RS-2024-00355370); and the Asan Institute for Life Sciences, Asan Medical Center, Seoul, Republic of Korea (2024IP0058). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. These authors contributed equally: Gil-Sun Hong, Namkug Kim.

Department of Biomedical Engineering, Brain Korea 21 Project, University of Ulsan College of Medicine, Asan Medical Center, Seoul, Republic of Korea Department of Convergence Medicine and Institute of Digital Healthcare, University of Ulsan College of Medicine, Seoul, Republic of Korea Department of Radiology, Gangneung Asan Hospital, University of Ulsan College of Medicine, Gangneung, Republic of Korea Jeju Halla General Hospital, Trauma Center, Jeju-do, Republic of Korea Department of Health Screening and Promotion Center, University of Ulsan College of Medicine, Asan Medical Center, Seoul, Republic of Korea Department of Radiology and Research Institute of Radiology, University of Ulsan College of Medicine, Asan Medical Center, Seoul, Republic of Korea Correspondence to Gil-Sun Hong or Namkug Kim. N.K. is a stockholder of Coreline Soft Co., Ltd. Coreline Soft Co., Ltd. and JLK Inc. provided access to their AI algorithms for use in this study; however, these companies had no role in the study design, data collection, analysis, decision to publish, or preparation of the manuscript.