Agatston-2.0: A Next-Generation AI-Based Coronary Calcium Quantification Approach to Improve Risk Stratification Among Individuals with Zero Agatston Scores – Part I

Figures

• 

  Fig. 1. Manuscript Exclusion Criteria Flow Chart. Abbreviations: FHS, Framingham Heart Study; MESA, Multi-Ethnic Study of Atherosclerosis.

• 

  Fig. 2. Schematic overview of the AI-assisted coronary artery calcium (Agatston-2.0) quantification framework. Step 1: Automated deep learning–based segmentation of the coronary artery tree, including the left main, left anterior descending, left circumflex, and right coronary arteries. Step 2: AI-assisted intensity calibration and noise normalization using physical phantom measurements when available, or patient-derived reference tissues when phantoms are absent. Step 3: Voxel-wise relative density weighting using a continuous probabilistic transformation to enhance sensitivity to low-burden and small calcifications, particularly in thin-slice CT acquisitions. Spatial near-neighbor filtering is applied to suppress noise and aggregate biologically coherent calcification, with the final AI-CAC score computed as the sum of all weighted voxels within the coronary artery mask. Right panel: Comparison of conventional Agatston-1.0 and Agatston-2.0 approaches. Traditional Agatston-1.0 relies on thick-slice CT (∼3.0 mm), which may miss small or low-density calcifications due to partial volume effects. In contrast, Agatston-2.0 leverages thin-slice imaging (∼0.6 mm) and probabilistic voxel-wise…

• 

  Fig. 3. Cumulative incidence of coronary heart disease stratified by AI-CAC status among participants with CAC= 0 at baseline.Kaplan–Meier curves show the cumulative incidence of coronary heart disease (CHD) during follow-up in (A) the Multi-Ethnic Study of Atherosclerosis (MESA), (B) the Framingham Heart Study (FHS; Generations 2 and 3), and (C) the pooled cohort. Analyses were restricted to participants with CAC = 0 at baseline and stratified by AI-CAC status (AI-CAC score = 0 vs AI-CAC score > 0). The dashed blue line represents all participants with CAC = 0, the green line represents participants with AI-CAC score = 0, and the red line represents participants with AI-CAC score > 0. Numbers of participants in each group are shown in the legends. P values were calculated using the log-rank test.

• 

  Fig. 4. Cumulative incidence of coronary heart disease stratified by tertiles of AI-CAC score among participants with CAC = 0 at baseline. Kaplan–Meier curves depict the cumulative incidence of coronary heart disease (CHD) during follow-up in (A) the Multi-Ethnic Study of Atherosclerosis (MESA), (B) the Framingham Heart Study (FHS; Generations 2 and 3), and (C) the pooled cohort. Analyses were restricted to participants with CAC = 0 at baseline and stratified by tertiles of the AI-CAC score (T1–T3), as well as by AI-CAC score = 0. The blue line represents participants with AI-CAC score = 0, the green line represents the lowest nonzero tertile (T1), the orange line represents the middle tertile (T2), and the red line represents the highest tertile (T3). Numbers of participants and events in each group are shown in the legends. P values were calculated using the log-rank test, including pairwise comparisons of T3 versus AI- CAC score = 0 and overall comparisons across groups.

• 

  Central illustration.

• 

  Fig. 5. Detection of subthreshold coronary calcification by Agatston 2.0.(A) Axial non-contrast CT image demonstrating no visually apparent calcification by conventional criteria. (B) Magnified view of the right coronary artery (RCA) showing a subtle semi-calcified plaque. (C) Highlighted focus demonstrating voxel-level detection of low-attenuation calcification (orange overlay). This lesion corresponds to a 6-voxel plaque with attenuation in the 100–120 HU range, below the conventional 130 HU threshold for Agatston scoring. Agatston 2.0 quantification yields an AI-CAC phantom score of 17.6, illustrating its ability to capture subthreshold, low-density calcification not detected by standard methods.