Coronary Calcium Scoring without Dedicated Noncontrast CT
2021; Radiological Society of North America; Volume: 302; Issue: 2 Linguagem: Inglês
10.1148/radiol.2021212586
ISSN1527-1315
Autores Tópico(s)Medical Imaging Techniques and Applications
ResumoHomeRadiologyVol. 302, No. 2 PreviousNext Reviews and CommentaryFree AccessEditorialCoronary Calcium Scoring without Dedicated Noncontrast CTJames W. Goldfarb , J. Jane CaoJames W. Goldfarb , J. Jane CaoAuthor AffiliationsFrom the Department of Research and Education, St Francis Hospital and Heart Center, 100 Port Washington Blvd, Roslyn, NY 11576.Address correspondence to J.W.G. (e-mail: [email protected]).James W. Goldfarb J. Jane CaoPublished Online:Nov 23 2021https://doi.org/10.1148/radiol.2021212586MoreSectionsPDF ToolsImage ViewerAdd to favoritesCiteTrack CitationsPermissionsReprints ShareShare onFacebookTwitterLinked In See also the article by Mu and Bai et al in this issueDr Goldfarb is a cardiovascular imaging scientist in the Cardiac Imaging Section at the St Francis Hospital and Heart Center in Roslyn, New York. His research interests focus on novel cardiovascular imaging technical developments, usage, and reimbursement.Download as PowerPointOpen in Image Viewer Dr Cao is the director of the Cardiac Imaging Section at the St Francis Hospital and Heart Center in Roslyn, New York, and clinical professor of medicine at Stony Brook University in Stony Brook, New York. Her research interests include all aspects of cardiac imaging, with a focus on MRI and CT.Download as PowerPointOpen in Image Viewer Comprehensive morphologic assessment of coronary arteries with CT is performed in two separate acquisitions, a noncontrast acquisition for the quantitative assessment of coronary artery calcium (CAC) followed by angiography performed during bolus injection of contrast material for the assessment of coronary anatomy, the coronary lumen, and detailed plaque features. The overall goals are to evaluate the burden of coronary atherosclerosis, assess the risk of cardiovascular events, and aid in personalizing interventional and medical therapies for both hospital and outpatient settings. Despite some reservations to routinely perform both coronary CAC scoring and CT angiography (CTA) (1), both components are increasingly represented positively in appropriateness criteria and clinical research. Therefore, routine use of both CAC scoring and CTA is common (2).The 2018 American Heart Association and American College of Cardiology Guideline on the Management of Blood Cholesterol recommends that when a provider is uncertain of the cardiovascular risk status, a CAC score is a beneficial option to facilitate decision making in adults aged 40 years or older. The 2019 American College of Cardiology and American Heart Association Guideline on the Primary Prevention of Cardiovascular Disease states that the assessment of CAC score is a reasonable tool to reclassify risk for many individuals with borderline or intermediate estimated 10-year risk and for those patients reluctant to undergo medical therapy. Additionally, CAC characteristics, such as scoring components (ie, volume, area, and density), may be emerging biomarkers of interest in statin users (3). Unfortunately, the addition of CAC scanning to CTA has several drawbacks. The elimination of the additional noncontrast examination is beneficial to achieve as low as reasonably achievable radiation exposure. Patient throughput also benefits by skipping an unnecessary examination component. The radiation dose from obtaining a CAC score is relatively low, but its relative contribution to the coronary artery imaging examination has increased because of advances in CTA. On the other hand, CAC imaging can be used to rule out CAC and to prescribe low tube voltage CTA so as to reduce radiation dose. The downside would be misclassification because of increasing coronary lumen attenuation and increasing incidence of isodense plaques (4).In this issue of Radiology, Mu and Bai et al (5) use recent technologic developments—spectral CT, a virtual noncontrast (VNC) reconstruction, and deep learning (DL), thus eliminating the need for a dedicated noncontrast CAC scoring acquisition for patients undergoing CTA. Several reports describe the use of DL to segment CAC and to provide a CAC score (6, 7). Mu and Bai et al (5) implemented direct use of the coronary CTA acquisition (2). The task requires segmentation and mapping of calcium from the contrast-enhanced CTA scan to traditional noncontrast CAC imaging. Mapping is necessary because of the different CTA and dedicated CAC acquisition parameters, such as section thickness and tube voltages, that could affect the resulting CAC score. Supervised DL typically requires annotated image sets for training. In the case of CAC scoring, DL requires segmentation of CAC on the contrast-enhanced CTA scans, use of misregistered segmented noncontrast scans, or use of only the CAC score number. Segmentation of CAC on the contrast-enhanced CTA scans is challenging for a human observer because of the similar and variable attenuation of the contrast-enhanced lumen and CAC. Use of misregistered segmented noncontrast scans would require complex three-dimensional registration because of successive breath holds, as well as varying temporal and spatial resolution, although breath-hold difference is not critical when a volume scan is employed. Use of only the CAC score number would also be suboptimal because of the exclusion of the plethora of information in the CTA acquisition. To overcome these obstacles, DL training was performed using spectral CTA acquisition, which allowed for the use of a VNC reconstruction for easy segmentation with perfect alignment to the CTA examinations. Use of spectral CT acquisition in training did not necessitate its use in testing or application, and any CTA acquisition, spectral or nonspectral, could be used to generate a CAC score. The authors additionally use two modules in their DL architecture—a CAC detection module and a CAC score regression module. The first detects CAC in the CTA acquisition using a three-dimensional convolutional neural network, and the second maps the variable signal intensities at CTA to a traditional CAC score. In the past, a simple linear regression was used, but DL allows for complex nonlinear mapping.An interesting aspect of the report by Mu and Bai et al (5) is the use of not only the annotated calcium CTA scans but also the CAC score from traditional noncontrast calcium scoring. Although Mu and Bai et al (5) report excellent correlation between VNC-based and traditional CAC scoring, the images differ, and biases in CAC score have been reported with acquisition protocol changes (8). The use of both images and the CAC score allowed the convolutional neural network in the regression module to map the final CAC score to traditional values.The authors follow good DL practice with separate training, validation, and testing sets, as well as evaluation of intra- and interobserver variability in comparison to the DL method. The model was developed using 365 patients and evaluated on an independent testing set of 240 patients. The conventional noncontrast and VNC CAC scores both had excellent observer agreement. The intraclass correlation coefficient (ICC) range was 0.96–1.0. Surprisingly, there was excellent correlation and agreement between conventional and VNC CAC scoring (Pearson correlation coefficient = 0.95; ICC = 1.0). This is unexpected as demonstrated visually in figure 3 of the article and reported in the literature. The exact protocols for traditional and VNC scoring were not detailed with respect to Hounsfield unit thresholds and reconstruction parameters such as section thickness. In addition, the patient characteristics were scarcely described. The most important of all is the body habitus, which can profoundly affect image quality of both the noncontrast and contrast-enhanced scans. The lack of patient information creates uncertainty about the generalizability of this report. Although the intraobserver variability was reportedly excellent, with an ICC of 0.96 (95% CI: 0.95, 0.97), the mean difference in the CAC score reported by two observers was surprisingly large at 40 ± 185 (± standard deviation), especially for images generated from conventional noncontrast imaging, which is expected to have superb reproducibility. Additionally, they tested their developed software on an independent test set (240 patients) of CTA scans and compared it to traditional noncontrast scoring, not just VNC scoring, which is not common in current practice. This includes both spectral CTA and nonspectral CTA, with the latter being more commonplace in practice.Validation on 240 independent CTA scans from three different CT scanners showed excellent correlation and agreement with CAC scores measured manually from dedicated CAC acquisitions during the same session (Pearson correlation coefficient = 0.96; ICC = 0.94). When placed into the four risk categories (CAC score: 0–10, 11–100, 101–400, or > 400), agreement was excellent (κ = 0.94) between DL CTA and traditional noncontrast CAC scores. In terms of risk assessment, CAC scoring according to CTA and according to semiautomatic noncontrast CT risk differed by at most one risk category in only 7% of the patients (17 of 240 patients). The report is encouraging but is primarily proof of concept. Coronary artery plaques have been reported to be missed at coronary CTA performed without concomitant CAC scanning (4). Unfortunately, Mu and Bai et al (5) did not study CAC detected with CTA with DL on a coronary vessel or CAC lesion basis. This is most likely necessary before clinical adoption. It is encouraging that the method was trained using examinations from a single spectral CT scanner and tested with several nonspectral CT scanners. Although generalization to scanners outside of the single hospital was not demonstrated, this limitation was acknowledged. Additionally, the number of patients used for training and evaluation is small. Last, this DL method yields only a total CAC score and not a vessel-specific score. A vessel-specific score would be useful for follow-up comparisons and for determining risk.The primary goal of CTA is the inclusion or exclusion of obstructive coronary artery disease, but increasingly the plaque features and burden become clinically important for risk stratification (9). Dedicated CAC scanning and score have historically been used for several reasons in CTA examinations, the most notable of which is perhaps the well-established prognostic value of CAC scores, which has predated that of CTA. Although there is a positive correlation between CAC and the presence and severity of obstructive coronary artery disease, it is highly debatable if CAC scoring is indicated in all patients undergoing CTA; nonetheless, CAC score remains increasingly important in primary prevention. In conclusion, this study has implications not only for reducing radiation dose and for improving patient throughput, but also for offering retrospective CAC scoring for patients when dedicated CAC score imaging is not, thus allowing for enhanced risk stratification and adherence to current statin guidelines.Disclosures of Conflicts of Interest: J.W.G. No relevant relationships. J.J.C. No relevant relationships.References1. Incze MA. Should I Get a Coronary CT Scan? JAMA Intern Med 2021;181(5):732. Crossref, Medline, Google Scholar2. Goldfarb JW, Weber J. Trends in Cardiovascular MRI and CT in the U.S. Medicare Population from 2012 to 2017. Radiol Cardiothorac Imaging 2021;3(1):e200112. Link, Google Scholar3. Osei AD, Mirbolouk M, Berman D, et al. Prognostic value of coronary artery calcium score, area, and density among individuals on statin therapy vs. non-users: The coronary artery calcium consortium. Atherosclerosis 2021;316(79):83. Google Scholar4. Baliyan V, Scholtz JE, Kordbacheh H, Hedgire S, Ghoshhajra BB. False-Negative Low Tube Voltage Coronary CT Angiography: High Intravascular Attenuation at Coronary CT Angiography Can Mask Calcified Plaques. Radiol Cardiothorac Imaging 2019;1(4):e190039. Link, Google Scholar5. Mu D, Bai J, Chen W, et al. Calcium scoring at coronary CT angiography using deep learning. Radiology 2022;302(2):309–316. Link, Google Scholar6. Takx RA, de Jong PA, Leiner T, et al. Automated coronary artery calcification scoring in non-gated chest CT: agreement and reliability. PLoS One 2014;9(3):e91239. Crossref, Medline, Google Scholar7. van Velzen SGM, Lessmann N, Velthuis BK, et al. Deep Learning for Automatic Calcium Scoring in CT: Validation Using Multiple Cardiac CT and Chest CT Protocols. Radiology 2020;295(1):66–79. Link, Google Scholar8. Nadjiri J, Kaissis G, Meurer F, et al. Accuracy of Calcium Scoring calculated from contrast-enhanced Coronary Computed Tomography Angiography using a dual-layer spectral CT: A comparison of Calcium Scoring from real and virtual non-contrast data. PLoS One 2018;13(12):e0208588. Crossref, Medline, Google Scholar9. Chang HJ, Lin FY, Lee SE, et al. Coronary Atherosclerotic Precursors of Acute Coronary Syndromes. J Am Coll Cardiol 2018;71(22):2511–2522. Crossref, Medline, Google ScholarArticle HistoryReceived: Oct 12 2021Revision requested: Oct 22 2021Revision received: Oct 25 2021Accepted: Oct 26 2021Published online: Nov 23 2021Published in print: Feb 2022 FiguresReferencesRelatedDetailsAccompanying This ArticleCalcium Scoring at Coronary CT Angiography Using Deep LearningNov 23 2021RadiologyRecommended Articles Calcium Scoring at Coronary CT Angiography Using Deep LearningRadiology2021Volume: 302Issue: 2pp. 309-316Detecting Coronary Artery Calcium on Chest Radiographs: Can We Teach an Old Dog New Tricks?Radiology: Cardiothoracic Imaging2021Volume: 3Issue: 3Deep Learning for Automatic Calcium Scoring in CT: Validation Using Multiple Cardiac CT and Chest CT ProtocolsRadiology2020Volume: 295Issue: 1pp. 66-79Coronary CT Angiography: Variability of CT Scanners and Readers in Measurement of Plaque VolumeRadiology2016Volume: 281Issue: 3pp. 737-748Incremental Prognostic Value of Coronary Artery Calcium Score for Predicting All-Cause Mortality after Transcatheter Aortic Valve ReplacementRadiology2021Volume: 301Issue: 1pp. 105-112See More RSNA Education Exhibits Coronary Artery Calcium Scoring: Past, Present and FutureDigital Posters2020Coronary Artery Calcium Scoring: Current Practice And Future DirectionsDigital Posters2021Clinical Applications of Artificial Intelligence in Cardiovascular ImagingDigital Posters2020 RSNA Case Collection Anomalous origin of the right coronary artery from the left sinus of ValsalvaRSNA Case Collection2020Coral Reef AortaRSNA Case Collection2022Caseous Mitral Annular Calcification RSNA Case Collection2022 Vol. 302, No. 2 Metrics Altmetric Score PDF download
Referência(s)