An artificial intelligence–based coronary artery calcium scoring system typically generated reports within 10 minutes of image upload after integration into routine clinical workflow at a tertiary hospital in South Korea, according to an editorial published in the Korean Journal of Radiology.
The editorial described implementation of an artificial intelligence (AI)–based coronary artery calcium (CAC) scoring platform at Wonkwang University Hospital, a university-affiliated tertiary hospital and regional cardiovascular and cerebrovascular center in Iksan, Republic of Korea. The institution performs approximately 170 electrocardiogram-gated cardiac computed tomography (CT) examinations per month, most involving CAC scoring, across multiple CT platforms from different vendors.
The authors said the institution has applied AI-based CAC scoring to all electrocardiogram-gated noncontrast cardiac CT examinations since April 2025. In Korea, AI-based CAC scoring systems received regulatory approval in July 2021.
The implemented system, AVIEW CAC version 1.1 from Coreline Soft, combines an atlas-guided approach with deep learning–based semantic segmentation to identify coronary calcifications while minimizing misclassification of noncoronary calcium. Automated CAC results were uploaded to the picture archiving and communication system, INFINITT G7 from INFINITT Healthcare, alongside the original CT data.
Reports included total and vessel-specific Agatston, volume, and mass scores, as well as age- and sex-adjusted percentile-based risk stratification derived from population-based reference cohorts.
AI-generated CAC reports were typically available in the picture archiving and communication system within 10 minutes of CT image upload, according to the authors. They noted that early implementation was accompanied by transmission-related delays and intermittent AI report delivery issues, but those issues were resolved with system stabilization.
In an institutional audit, the authors said they observed a high level of agreement between AI-based CAC scoring and manual reference standards for continuous scores and categorical grading, particularly for clinically relevant thresholds. In the discussion, they highlighted CAC scores of 0 and CAC scores of at least 100, which can influence statin initiation decisions under current guidelines.
However, the editorial did not provide detailed audit methods, sample size, patient characteristics, how the manual reference standard was adjudicated, agreement statistics, sensitivity, specificity, error rates, or downstream clinical outcomes. The report therefore represents a single-center implementation experience rather than a formal validation or comparative effectiveness study.
Before implementation, all cardiac CT examinations at the institution were interpreted by a single radiologist, while CAC analysis was traditionally performed by radiographers using semiautomated software requiring manual region-of-interest placement and vessel labeling. The authors described that workflow as time-consuming and operator-dependent, with frequent reanalysis and workflow disruption.
Following implementation, AI-based CAC scoring replaced radiographer-generated preliminary reports, which the authors said reduced repetitive manual tasks, unnecessary reanalysis, and operator-dependent variability.
The authors also described limitations that remained after implementation. In their institutional experience, false-positive and false-negative detections persisted, most commonly related to noncoronary calcifications and motion artifacts. They also reported occasional vessel misclassification between the left main coronary artery and the left anterior descending artery in selected patients.
The authors noted that these limitations were more pronounced among patients with high CAC burden or structural coronary abnormalities, including coronary dilatation, occlusion, or anatomic variants. They said false-positive and false-negative detections have also been reported in the broader literature and are most commonly related to noncoronary calcifications, motion artifacts, excessive CAC burden, and complex coronary anatomy.
One example included in the editorial showed false-negative detection and vessel misclassification in a patient with coronary artery ectasia and intraluminal thrombus. In that case, some left circumflex artery calcifications were incorrectly classified as left anterior descending artery calcifications, while other hyperattenuating foci were identified as at least 130-Hounsfield-unit pixels but were not classified as CAC.
The authors emphasized that radiologist oversight remains essential, particularly when patients have extensive calcification or complex coronary anatomy. They also noted that transparent visualization and editable AI outputs are important so radiologists can efficiently verify and correct CAC results within routine workflows.
The editorial focused on electrocardiogram-gated noncontrast cardiac CT, the reference standard for CAC scoring. The authors distinguished this use from AI-based CAC assessment on non-electrocardiogram-gated chest CT, where standardized and robust quantitative approaches remain less established.
“Despite its technical maturity, radiologist oversight remains essential, particularly in patients with a high CAC burden or complex coronary anatomy, where false-positive or false-negative findings may still occur,” wrote Se Ri Kang, of Wonkwang University Hospital, Wonkwang University School of Medicine in Iksan, Republic of Korea, and Ji Young Rho, MD, PhD.
Disclosures: The authors reported no potential conflicts of interest.
Source: Korean Journal of Radiology