Table 1.
Deep learning techniques for automating CAC from gated scans.
| Study | Year Published | Study Size (Testing) | Algorithm Type | Risk Categories | Accuracy | Conclusions |
|---|---|---|---|---|---|---|
| Eng et al. [38] | 2021 | 79 | CNN | Categories for Agatston scores of 0, 1–10, 11–100, 101–400, >400 | Mean difference scores: −2.86, Kappa = 0.89, p < 0.0001 | Demonstrated near-perfect agreement with the reference standard and with improved computational speed. |
| Hong et al. [37] | 2022 | 959 | U-Net (CNN) | Categories for Agatston scores of 0, 1–10, 11–100, 101–400, >400 | ICC = 1.00, Kappa = 0.931 | Demonstrated excellent agreement with the reference standard and detected mild calcifications not detected by reference. |
| Gogin et al. [45] | 2021 | 98 | U-Net (CNN) | Categories for Agatston scores of 0, 1–10, 11–100, 101–400, >400 | Concordance-index = 0.951 | With an ensemble of 5 CNN models, there is high concordance with the standard reference. |
| Zhang et al. [58] | 2021 | 46 | U-Net (CNN) | Risk categorization not compared | ICC = 0.988, mean difference scores: −6.7, p = 0.993 | High-speed and accurate automated quantification of total and vessel-specific CAC in a single-center study. |
| Sandstedt et al. [54] | 2020 | 315 | CNN | Categories for Agatston scores of 0, 1–10, 11–100, 101–400, >400 | Mean difference scores: −8.2, ICC = 0.996, Kappa = 0.919 | Single-center study demonstrating near-perfect agreement, including Agatson assessment, volume score, mass score, and number of calcified lesions. |
| Wang et al. [59] | 2019 | 140 | 3D CNN | Categories for Agatson scores of 0, 1–99, 100–299, and >300 | ICC = 0.94, Kappa = 0.77 | Single-center study with near-perfect agreement of Agatson, volume, and mass scores and a reclassification rate of 13%. |
| Martin et al. [55] | 2020 | 511 | ResNet CNN | Categories for Agatston scores of 0, 1–10, 11–100, 101–400, >400 | ICC = 0.985, Kappa = 0.932 | Demonstrated outstanding agreement of total Agatson score with the reference standard trained on a dataset of 2000 patients. |
| Winkel et al. [57] | 2022 | 1171 | CNN | Categories for Agatston scores of 0, 1–10, 11–100, 101–400, >400 | ICC = 0.84, Kappa = 0.91 | Large, multicenter study demonstrating excellent accuracy on a total and per-vessel basis. |
| Idhayid et al. [39] | 2022 | 1849 | 3D CNN | Categories for Agatston scores of 0, 1–10, 11–100, 101–400, >400 | ICC = 0.98, Kappa = 0.90, p < 0.001 | Large study with scans obtained from multiple vendors demonstrated excellent agreement and efficiency. |