Advertisement

Journal of Nuclear Cardiology

, Volume 25, Issue 6, pp 2144–2147 | Cite as

Coronary calcium scoring of CT attenuation correction scans: Automatic, manual, or visual?

  • Shifali Dumeer
  • Andrew J. Einstein
Editorial
  • 501 Downloads

The advent of hybrid techniques combining myocardial perfusion imaging (MPI) and low-dose CT for attenuation correction has generated interest in the potential to obtain complimentary information regarding coronary artery calcium (CAC) from the CT component of the exam. The feasibility of this CT attenuation correction (CTAC) scan in identifying and quantifying coronary artery atherosclerotic calcifications has been evaluated in a few studies performed over the last decade, work aiming towards obviating the need for a dedicated coronary artery calcium scoring CT (CSCT), which contributes to study time, cost, and radiation burden.1-4 Estimated effective radiation dose from a CSCT varies widely depending on CT protocol, with values around 2 mSv being common,5 whereas dose from CTAC scan can be as low as 0.3 mSv.4 In addition, recently published joint guidelines from the Society of Cardiovascular Computed Tomography and the Society of Thoracic Radiology advocate reporting of CAC scores for cardiovascular disease (CVD) risk classification on all non-contrast CT examinations as the appropriate standard of care.6 Following these recommendations by manually performing calcium scoring would be a time-consuming and tedious process.

In prior studies, CAC has either been quantified manually or estimated visually. Two studies2,3 have focused on manual quantification of CAC from CTAC scans, utilizing commercially available software packages to place regions of interest on structures thought to represent coronary calcification, and determine CAC score with predefined thresholds. Both studies compared Agatston scores from CSCT to CAC score quantified on CTAC images at various attenuation thresholds for calcium. Kaster et al2 stated that they optimized the CTAC images for quantifying CAC by acquiring a low-dose, prospective ECG-gated CT scan, acquired with two times the standard CTAC tube current, at end-expiration breath-hold, and at 70% of the cardiac phase. For the CTAC scan, Mylonas et al3 studied thresholds of 50, 75, 100, 130 HU and Kaster et al investigated thresholds of 100, 110, 120, 130 HU. In both studies, patients were classified in CAC categories utilizing a four-level scale (0, 1-100, 101-400, >400). In a cohort of 23 patients, Kaster et al found highest correlation (r2 = 0.99) at a threshold of 110 HU, with CTAC underestimating calcium by less than 5%. In 91 patients, Mylonas et al found excellent agreement between CSCT and CTAC with highest agreement obtained at 50 HU with 79% of CTAC calcium scores concordant within the same category and the remainder within one category.

Other studies have focused on visual estimation of CAC. In a multicenter study including 492 patients, Einstein et al,4 evaluated the accuracy and reproducibility of visual estimation of CAC from low-dose CTAC scans performed for PET and SPECT MPI, compared to quantitative Agatston scoring in CSCT. Visually estimated CAC on CTAC images was applied to classify patients on a six-level scale as estimated AS of 0, 1-9, 10-99, 100-300, 400-999, or ≥1,000. The results revealed a high degree of association between visually estimated CAC and Agatston score with 63% of visually estimated scores in the same category as the Agatston score category and 93% varying by no more than one category, quadratic weighted kappa being 0.89. In a similar study of 250 patients who underwent SPECT/CT, Engbers et al1 found that visual estimation in CTAC correctly classified 71% of the CAC scores in the same category, and 94% within one category.

In this issue of Journal of Nuclear Cardiology®, Isgum et al for the first time investigated the potential of an automatic method for CAC scoring and CVD risk assessment applied to CTAC images acquired during cardiac PET/CT, with comparisons made to reference dedicated CSCT. The authors analyzed an in-house developed automated CAC scoring method based on intensity thresholding, statistical location of coronary arteries, and geometric features (volume, shape, intensity). The intensity threshold chosen was 130 HU. The study included 133 patients who underwent myocardial perfusion Rb-82 PET/CT. PET scanning was performed with CTAC, first at rest, and then with stress, followed by dedicated CSCT. Manual and automated CAC scoring was performed on all three CT scans, with manual CAC scoring in both CTAC and CSCT providing the reference standards. A five-level scale for risk categorization was used (0, 1-10, 11-100, 101-400, or >400). Agatston scores in CSCT and CTAC were significantly different at rest and stress and for automatic and manual CAC scoring due to different image acquisition parameters. These differences were compensated for by ranking the CAC scores in CTAC scans based on the reference CSCT risk category. The same number of patients was assigned to each group in CTAC scans as inferred by their ranking in the CSCT categorization. The automatic method correctly classified 70% of patients on stress CTAC images and 71% on rest images in the same CVD risk category, and 94% and 95% within one category, respectively. Linearly weighted κ values were 0.70 and 0.74, and two-way ICC values for absolute agreement were 0.70 and 0.68, respectively. The manual method correctly classified 76% of patients on stress CTAC images and 77% on rest images in the same CVD risk category, and 98% and 99% within one category, respectively. Linearly weighted κ were 0.79 and 0.82, and two-way ICC values for absolute agreement were 0.87 and 0.86, respectively.

An automatic calcium scoring method can alleviate the workload and interpretation time in a busy clinical practice. The important question at this time is whether automatic CAC method achieves a better combination of greater accuracy and improved workflow when compared to manual and visual CAC scoring methods. The visual method by Einstein et al4 resulted in 63% of patients with concordant scores and 93% of patients with scores within one risk category, with linearly weighted kappa of 0.77. These results are not much different from those achieved by Isgum et al. However, Einstein et al used a six-level scale for CVD risk categorization and Isgum et al utilized a five-level scale, limiting exact comparison. Also on analysis of the current study, manual scoring in CTAC was found to be more accurate than automatic scoring; there was 76%-77% accuracy with the manual method compared to 70%-71% accuracy with the automatic method. It is of great clinical importance to distinguish patients without any calcium from those with positive scores as even low CAC scores (1-10) are associated with significantly increased risk for incident cardiovascular events relative to those with scores of 0.7,8 In the study by Isgum et al, 11 of 24 (46%) and 9 of 20 (45%) of patients were classified by the automatic method as having no calcium on rest and stress CTAC, respectively, but actually had calcium on CSCT. In the same study, the manual method had lower false negative rates of 29% and 35% on rest and stress CTAC. False negative rates were also lower in prior studies evaluating visual and manual CAC scoring methods, being 17% for a manual method by Mylonas et al,3 and 22% and 25%-30% for visual methods by Einstein et al4 and Engbers et al,1 respectively. The automatic method occasionally missed large CAC lesions affected by cardiac motion or partial volume effect resulting in low contrast with the surrounding tissue. In fact, work from the same authors shows that automatic methods are prone to erroneous underestimation of calcium in the distal coronary artery and overestimation of aortic calcifications in coronary ostium.9 Perhaps, visual or manual estimation by an expert reader would be better able to assess these outliers. On the other hand, lesser interobserver variability would be expected with the automatic method when compared to visual method with the latter being a subjective assessment with increased variability expected among less experienced readers. It would be interesting to compare visual, manual, and automatic CAC scoring in the same study with the same set of data and investigators to get an accurate comparison of between-method agreement and interobserver reproducibility.

The time taken by the automatic software in calculation of CAC score, of approximately 20 minutes,9 occurred in a research context. This lengthy addition would not be practical for clinical workflows, whereas visual estimation by an experienced reader can be performed in a matter of seconds. Further work needs to be done to optimize the automatic software clinically by decreasing the computation times. Also for workflow, the automatic method should ideally be able to integrate with picture archiving and communications system (PACS) and reporting systems, with the calcium score calculation beginning as soon as the study is sent to the PACS and then results being automatically populated in the report.

A number of factors affect the CAC scores in CTAC compared to CSCT. Most of the factors lead to underestimation, including blurring of coronary arteries and gaps in the image data by cardiac motion artifact, partial volume effects by slice thickness selection, and low-resolution images produced as result of lower tube current and/or potential on CTAC. Factors leading to overestimation are high noise levels on low-dose CTAC, slice duplications produced by cardiac motion, and erroneous detection of calcium other than coronary calcification (metal implants and valve calcifications). A majority of these factors lead to underestimation as opposed to overestimation, resulting in lower CAC scores on CTAC compared to dedicated CSCT.10-12 Most of these factors are inherent to the CTAC technique and would lead to outliers irrespective of the CAC scoring method.

There are variable data in the literature about the optimal HU threshold that should be used for CAC quantification in CTAC, with some studies suggesting that use of the standard threshold of 130 HU may underestimate coronary calcification on low-dose CTAC, by missing lower attenuating calcific plaque due to increased image blurring and resultant lower contrast. The authors chose a 130 HU threshold as they observed that with their automatic method, CAC scoring using lower thresholds erroneously connected coronary calcifications to local bony structures. The ICC between CTAC and CSCT using manual scoring was 0.86-0.87. Mylonas et al3 reported comparable ICC of 0.80 using a threshold of 130 HU, however, obtained best results using a threshold of 50 HU with ICC of 0.95. Kaster et al2 found best agreement using a threshold of 110 HU and reported 20% and 5% underestimation of CAC at 130 and 110 HU thresholds. Einstein et al4 suggested that lower CT numbers up to 80 HU represent CAC. However, the findings of Wu et al,13 who observed that a standard Agatston method (threshold 130 HU) could be reliably used on low-dose CT scans performed for lung cancer screening, concur with those of the current author.

In contrast to Wu et al, Isgum et al acquired their reference CSCT at 120 kVp and their CTAC scan at 100 kVp, but used an identical threshold of 130 HU to compute CAC on both the scans. A tube potential of 100 kVp is associated with higher HU of calcifications than a tube potential of 120 kVp, and the lower potential will generally increase the CAC score. Nakazato et al,10 from the same Cedars-Sinai Group as Isgum et al, demonstrated that a low-radiation dose protocol using a tube potential of 100 kV for CSCT gives equivalent quantitative calcium results when compared to a standard CSCT method at 120 kVp, when the threshold at 100 kV is increased from 130 to 147 HU. This reference raises question whether 100 kVp CTAC should be scored with a greater threshold than CSCT. However, there are conflicting factors associated with CTAC that reduce the CAC score, like motion and lower resolution. Further research needs to determine the best HU threshold for computing CAC scores on CTAC images obtained with various scan parameters.

A limitation discussed by the authors is that they performed whole-heart calcium scoring and not comparison between individual coronary calcifications on CTAC and CSCT. Left main and left anterior descending coronary arteries have shown to have a higher prognostic importance than other coronary arteries.14 Qian et al demonstrated superior diagnostic accuracy of vessel- and lesion-specific CAC score when compared to whole-heart CAC score in predicting obstructive CAD.15 Accurate segmentation of coronary arteries on non-contrast CT images remains a challenging task and has mostly been done in literature with a concurrent contrast enhanced scan relying on the spatial information of coronary arteries extracted from the contrast cardiac CT image which outlines the coronary artery.16 Shahzad et al17 and Wolternik et al18 have demonstrated feasibility of CAC on non-contrast cardiac CT making use of statistical location information of coronary arteries extracted from a training set of contrast cardiac CT images to estimate the location of coronary arteries in the non-contrast images. Repeating a similar algorithm on a CTAC would be challenging owing to degradation of images by partial volume effect and cardiac motion. Further work needs to be done to optimize the software in order to obtain automatic individual-vessel CAC scores.

In conclusion, the study by Isgum et al raises the intriguing prospect of an automatic CAC scoring method which has the potential to be applied to a variety of non-gated non-contrast chest CT scans performed for different indications. The automatic software is not quite ready for prime time until it achieves accuracy similar to and workflow no worse than manual and visual scoring. Further, it needs to be validated in multicenter settings over different vendor acquisitions. Nevertheless, this work by Isgum et al is an important step in the development of automated methods for CAC scoring which can be applied to millions of non-gated non-cardiac studies, providing additional information for risk stratification of CAD.

Notes

Disclosures

Dr. Einstein is supported in part by Grant R01 HL10971 from the National Heart Lung and Blood Institute, and has received Research Grants to Columbia University from GE Healthcare, Philips Healthcare, and Toshiba America Medical Systems. No disclosures for the author Shifali Dumeer.

References

  1. 1.
    Engbers EM, Timmer JR, Mouden M, Jager PL, Knollema S, Oostdijk AH, et al Visual estimation of coronary calcium on computed tomography for attenuation correction. J Cardiovasc Comput Tomogr 2016;10:327-9.CrossRefGoogle Scholar
  2. 2.
    Kaster TS, Dwivedi G, Susser L, Renaud JM, Beanlands RS, Chow BJ, et al Single low-dose CT scan optimized for rest–stress PET attenuation correction and quantification of coronary artery calcium. J Nucl Cardiol Off Publ Am Soc Nucl Cardiol 2015;22:419-28.CrossRefGoogle Scholar
  3. 3.
    Mylonas I, Kazmi M, Fuller L, deKemp RA, Yam Y, Chen L, et al Measuring coronary artery calcification using positron emission tomography–computed tomography attenuation correction images. Eur Heart J Cardiovasc Imaging 2012;13:786-92.CrossRefGoogle Scholar
  4. 4.
    Einstein AJ, Johnson LL, Bokhari S, Son J, Thompson RC, Bateman TM, et al Agreement of visual estimation of coronary artery calcium from low-dose CT attenuation correction scans in hybrid PET/CT and SPECT/CT with standard Agatston score. J Am Coll Cardiol 2010;56:1914-21.CrossRefGoogle Scholar
  5. 5.
    Kim KP, Einstein AJ, Berrington de Gonzalez A. Coronary artery calcification screening: Estimated radiation dose and cancer risk. Arch Intern Med 2009;169:1188-94.CrossRefGoogle Scholar
  6. 6.
    Hecht HS, Cronin P, Blaha MJ, Budoff MJ, Kazerooni EA, Narula J, et al 2016 SCCT/STR guidelines for coronary artery calcium scoring of noncontrast noncardiac chest CT scans: A report of the Society of Cardiovascular Computed Tomography and Society of Thoracic Radiology. J Cardiovasc Comput Tomogr 2017;11:74-84.CrossRefGoogle Scholar
  7. 7.
    Blaha M, Budoff MJ, Shaw LJ, Khosa F, Rumberger JA, Berman D, et al Absence of coronary artery calcification and all-cause mortality. JACC Cardiovasc Imaging 2009;2:692-700.CrossRefGoogle Scholar
  8. 8.
    Budoff MJ, McClelland RL, Nasir K, Greenland P, Kronmal RA, Kondos GT, et al Cardiovascular events with absent or minimal coronary calcification: The Multi-Ethnic Study of Atherosclerosis (MESA). Am Heart J 2009;158:554-61.CrossRefGoogle Scholar
  9. 9.
    Wolterink JM, Leiner T, de Vos BD, Coatrieux JL, Kelm BM, Kondo S, et al An evaluation of automatic coronary artery calcium scoring methods with cardiac CT using the orCaScore framework. Med Phys 2016;43:2361. doi: 10.1118/1.4945696.CrossRefGoogle Scholar
  10. 10.
    Nakazato R, Dey D, Gutstein A, Le Meunier L, Cheng VY, Pimentel R, et al Coronary artery calcium scoring using a reduced tube voltage and radiation dose protocol with dual-source computed tomography. J Cardiovasc Comput Tomogr 2009;3:394-400.CrossRefGoogle Scholar
  11. 11.
    Muhlenbruch G, Thomas C, Wildberger JE, Koos R, Das M, Hohl C, et al Effect of varying slice thickness on coronary calcium scoring with multislice computed tomography in vitro and in vivo. Investig Radiol 2005;40:695-9.CrossRefGoogle Scholar
  12. 12.
    Horiguchi J, Fukuda H, Yamamoto H, Hirai N, Alam F, Kakizawa H, et al The impact of motion artifacts on the reproducibility of repeated coronary artery calcium measurements. Eur Radiol 2007;17:81-6.CrossRefGoogle Scholar
  13. 13.
    Wu MT, Yang P, Huang YL, Chen JS, Chuo CC, Yeh C, et al Coronary arterial calcification on low-dose ungated MDCT for lung cancer screening: Concordance study with dedicated cardiac CT. Am J Roentgenol 2008;190:923-8.CrossRefGoogle Scholar
  14. 14.
    Williams M, Shaw LJ, Raggi P, Morris D, Vaccarino V, Liu ST, et al Prognostic value of number and site of calcified coronary lesions compared with the total score. JACC Cardiovasc Imaging 2008;1:61-9.CrossRefGoogle Scholar
  15. 15.
    Qian Z, Anderson H, Marvasty I, Akram K, Vazquez G, Rinehart S, et al Lesion- and vessel-specific coronary artery calcium scores are superior to whole-heart Agatston and volume scores in the diagnosis of obstructive coronary artery disease. J Cardiovasc Comput Tomogr 2010;4:391-9.CrossRefGoogle Scholar
  16. 16.
    Yang G, Chen Y, Ning X, Sun Q, Shu H, Coatrieux JL. Automatic coronary calcium scoring using noncontrast and contrast CT images. Med Phys 2016;43:2174. doi: 10.1118/1.4945045.CrossRefGoogle Scholar
  17. 17.
    Shahzad R, van Walsum T, Schaap M, Rossi A, Klein S, Weustink AC, et al Vessel specific coronary artery calcium scoring: An automatic system. Acad Radiol 2013;20:1-9.CrossRefGoogle Scholar
  18. 18.
    Wolterink JM, Leiner T, Takx RA, Viergever MA, Isgum I. Automatic coronary calcium scoring in non-contrast-enhanced ECG-triggered cardiac CT with ambiguity detection. IEEE Trans Med Imaging 2015;34:1867-78.CrossRefGoogle Scholar

Copyright information

© American Society of Nuclear Cardiology 2017

Authors and Affiliations

  1. 1.Department of RadiologyColumbia University Medical Center and New York-Presbyterian HospitalNew YorkUSA
  2. 2.Department of Medicine, Cardiology Division, and Department of RadiologyColumbia University Medical Center and New York-Presbyterian HospitalNew YorkUSA

Personalised recommendations