Introduction
Coronary artery calcium score (CACS)
Technical parameters | |
Patient position | Supine |
Type of acquisition | Axial |
Scanning mode | Prospective, ECG-gated |
Scan range | From below the aortic arch to the base of the heart |
R-R interval | 70-80% |
Slice thickness | 2.5-3 mm |
Reconstruction thickness | 2.5-3 mm |
Peak tube voltage | 120 kV |
Tube current | Modulated current based on BMI |
Reconstruction algorithm | Filtered back projection |
Reconstructed matrix | 512 X 512 |
Patient preparation | |
Dietary preparation | Not needed |
Pharmacological therapy | No changes |
β-blockers | Not mandatory |
Contrast medium | Not administered |
Clinical value of CACS
Guideline | Population | Role of CACS | Reference |
---|---|---|---|
American college of cardiology / American Heart Association | Adults (40 to 75 years of age) at intermediate (7.5-20%) or selected adults at borderline (5-7.5%) 10-year ASCVD risk with uncertain risk-based decision for preventive therapy | CACS = 0 consider withhold statin (unless diabetes, smoking, etc.) CACS 1-99 favor statin (particularly, >55 years of age) CACS >100 or >75th percentile start statin therapy | [116] |
Younger or older adults (<45 or ≥75 years of age, respectively), women at lower 10-year ASCVD risk (<7.5%), or selected low-risk adults (<5%) 10-year ASCVD risk | Refine ASCVD risk | ||
Canadian Cardiovascular Society$ | Adults (>40 years of age) at intermediate (10-19.9%) 10-year FRS risk with an uncertain risk-based decision for preventive therapy | Initiate statin therapy if CACS >0 | [117] |
Selected adults (>40 years of age) at low (<10%) 10-year FRS risk (familiar history of premature ASCVD) | Initiate statin therapy if CACS >0 | ||
European Society for Cardiology / European Atherosclerosis Society | Asymptomatic individuals at low or moderate ASCVD risk who are eligible for statin therapy or those who were not able to lower cholesterol levels with lifestyle intervention alone | CACS >100 determines upward risk reclassification, leading to considering statins use | [118] |
United Kingdom National Institute for Health and Care Excellence | Asymptomatic patients with suggested electrocardiographic changes for ischemia | Adjudication statin allocation | [15] |
Cardiac Society of Australia and New Zealand% | Asymptomatic adults (45 to 75 years of age) at intermediate risk score (10-20%) for 10-year ASCVD | CACS = 0 no treatment CACS 1-100 improve diet and lifestyle changes CACS 100-400 aspirin recommended; statin considered reasonable CACS >400 statin and aspirin recommended | [119] |
Patients at lower 10-year risk (6-10%) with a strong family history of premature ASCVD or diabetic patients aged 40 to 60 years old | |||
Japanese Atherosclerosis Society+ | Not included in the predictive model. However, it is regarded as a prognostic tool in intermediate-to-high-risk individuals | Further high-quality studies, specifically addressing the Japanese population, are needed | [120] |
Chinese Society of Cardiology& | Not included in the predictive model. However, it may be used as a cardiovascular disease risk enhancement factor in patients aged 40 to 70 years of age who are at very high 10-year ASCVD risk | No role for CACS = 0 CACS ≥ 100 may trigger low-dose aspirin administration | [121] |
National Lipid Association | Adults (40 to 75 years of age) at borderline- to intermediate-risk (LDL-C 70 to 189 mg/dL and 5-19.9% 10-year ASCVD risk) | CACS = 0 defer statin CACS 1-99 favor statin CACS 100-299 favor statin and aspirin CACS ≥ 300 high-intensity statins and aspirin | [122] |
Adults (40 to 75 years of age) with LDL-C 70 to 189 mg/dL and <5% 10-year ASCVD risk and family history of premature ASCVD | CACS = 0 lifestyle changes CACS >0 lifestyle changes and consider statins | ||
Society of Cardiovascular Computed Tomography | Adults (40 to 75 years of age) with a 5% to 20% 10-year ASCVD risk group | CACS = 0 statin not recommended CACS 1-99 moderate or moderate to high intensity statin CACS 100-299 moderate to high intensity statin and aspirin CACS >300 high intensity statin and aspirin | [23] |
Adults (40 to 75 years of age) with a <5% 10-year ASCVD risk who seek reassurance (ex. strong family history of premature CAD, etc.) | CACS = 0 or low values confirm the low-risk status Higher CACS identify individuals in whom lifestyle recommendations should be enhanced or treatment considered |
CACS in non-ECG-gated images
Limits and future directions of CACS
Artificial intelligence
AI performance assessment
Metrics | Formula |
---|---|
Accuracy |
\(\frac{{TP + TN}}{{TP + TN + FP + FN}}\) |
Sensitivity (or recall) |
\(\frac{{TP}}{{TP + FN}}\) |
Specificity |
\(\frac{{TN}}{{FP + TN}}\) |
Precision |
\(\frac{{TP}}{{TP + TN}}\) |
F1 score |
\(2 \times \frac{{{\text{Precision}} \times {\text{Recall}}}}{{{\text{Precision}} + {\text{Recall}}}}\) |
Area under the receiving operator |
\(\frac{{S_{p} - \frac{{n_{p} \left( {n_{n} + 1} \right)}}{2}}}{{n_{p} n_{n} }}\) |
Jaccard index |
\(\frac{{\left| {X \cap Y} \right|}}{{\left| X \right| + \left| Y \right| + \left| {X \cap Y} \right|}}\) |
DICE score |
\(2\frac{{\left| {X \cap Y} \right|}}{{\left| X \right| + \left| Y \right|}}\) |
Translating AI concepts into CACS
CAC detection and segmentation
-
Calcified plaque detection based on the localization of large structures (i.e., cardiac profile and the aortic root). This approach allowed either image co-registration with previously built atlases, deriving the expected location of coronary arteries [77], or isolating the heart by applying various subsequent segmentation steps. Subsequently, calcifications were identified on the segmented images by image thresholding or geometrical constraints locating coronary arteries’ origin [78, 79].
-
ML-based selection of imaging features correctly classifying the presence of CAC [75, 76]. In this scenario, different approaches were explored, ranging from those necessitating user inputs [75, 80] to fully automatized ones [76]. At their core, these approaches rely on letting software grow regions of interest from which different features were derived. Features were further subdivided into intensity-based features (i.e., mean or maximum density), spatial features (i.e., the cartesian coordinates of the plaque), or geometrical features (i.e., the shape and size of the plaque) [81]. The best feature combination, enabling accurate CAC detection, was calculated by combining and testing various models [75, 76, 80]. Interestingly, lesion location and plaque highest density always led to the best model performances, whereas shape- and dimension-related features were consistently discharged [75, 76]. The results of these approaches varied considerably, with a study reporting a calcification detection of ~ 74% [76], while others had sensitivity and specificity values of 92–93% and 98–99%, respectively [75].
-
ML-based derivation of imaging features obtained using coronary arteries-based atlases, created upon CT-angiography images [74, 81, 82]. This method summed up the advantages of the formers using both lesion features and atlases and yielded CAC detection sensitivity between 81 and 87%, while specificity varied between 97% and 100% [82].
CACS quantification
First Author | Year of publication and reference | Structure of the algorithm | Dataset - Type of images | Number of patients | Age of patients in years, mean ± SD (or age range) | Performance | ||||
---|---|---|---|---|---|---|---|---|---|---|
Metric | LM | LAD | LCx | RCA | ||||||
Winkel et al. | 2022 [84] | 3-D U-net | ECG-gated CT | 1171 | 56±10 (510 pts) 58±9 (399 pts) 60±10 (262 pts) | Accuracy | 89% | 91% | 93% | 100% |
Hong et al. | 2022 [91] | U-net++ | ECG-gated CT | 1811 | 58 (18 to 96*) | Detection rate | 80% | 97% | 89% | 94% |
Zhang et al. | 2021 [86] | 3-D U-net | ECG-gated CT | 232 | 55±13 | ICC | 0.98 | 0.99 | 0.97 | 0.98 |
Morf et al. | 2022 [95] | 3-D U-net | PET/CT derived images | 100 | 66±11 | Accuracy | 79% | 87% | 75% | 79% |
Lesserman et al. | 2018 [74] | Double CNN | Low-dose chest CT | 1744 | 55 to 74* | Sensitivity | 93 %$ | 72 % | 92 % | |
Sartoretti et al. | 2023 [85] | 3-D U-net | ECG-gated CT | 56 | 63±9 | ICC | 0.64 | 0.95 | 0.93 | 0.99 |
Takahashi et al. | 2023 [123] | 3-D U-net | ECG-gated CT | 1369# | 63±13 (500 pts) 66±12 (409 pts) 60±10 (400 pts) | Pearson’s correlation | 0.85 | 0.98 | 0.98 | 0.99 |