Fully automated derivation of coronary artery calcium scores and cardiovascular risk assessment from contrast medium-enhanced coronary CT angiography studies

Deep Learning-Based Automated Quantification of Coronary Artery Calcification for Contrast-Enhanced Coronary Computed Tomographic Angiography

Background: We evaluated the accuracy of a deep learning-based automated quantification algorithm for coronary artery calcium (CAC) based on enhanced ECG-gated coronary CT angiography (CCTA) with dedicated coronary calcium scoring CT (CSCT) as the reference. Methods: This retrospective study included 315 patients who underwent CSCT and CCTA on the same day, with 200 in the internal and 115 in the external validation sets. The calcium volume and Agatston scores were calculated using both the automated algorithm in CCTA and the conventional method in CSCT. The time required for computing calcium scores using the automated algorithm was also evaluated. Results: Our automated algorithm extracted CACs in less than five minutes on average with a failure rate of 1.3%. The volume and Agatston scores by the model showed high agreement with those from CSCT with concordance correlation coefficients of 0.90–0.97 for the internal and 0.76–0.94 for the external. The accuracy for classification was 92% with a 0.94 weighted kappa for the internal and 86% with a 0.91 weighted kappa for the external set. Conclusions: The deep learning-based and fully automated algorithm efficiently extracted CACs from CCTA and reliably assigned categorical classification for Agatston scores without additional radiation exposure.

Feasibility and accuracy of coronary artery calcium score on virtual non-contrast images derived from a dual-layer spectral detector CT: A retrospective multicenter study

Rationale and objective

This retrospective study was to evaluate the feasibility and accuracy of coronary artery calcium score (CACS) from virtual non-contrast (VNC) images in comparison with that from true non-contrast (TNC) images.

Materials and methods

A total of 540 patients with suspected of coronary artery disease (CAD) who underwent a dual-layer spectral detector CT (SDCT) in three hospitals were eligible for this study and 233 patients were retrospectively enrolled for further analysis. The CACS was calculated from both TNC and VNC images and compared. Linear regression analysis of the CACS was performed between TNC and VNC images.

Results

The correlation of overall CACS from VNC and TNC images was very strong (r = 0.923, p < 0.001). The CACS from VNC images were lower than that from TNC images (221 versus. 69, p < 0.001). When the regression equation of the overall coronary artery was applied, the mean calibrated CACS-VNC was 221 which had a significant difference from the CACS-TNC (p = 0.017). When the regression equation of each coronary branch artery was applied, the mean calibrated CACS-VNC was 221, which had a significant difference from the CACS-TNC (p = 0.003). But the mean difference between the CACS-TNC and the calibrated CACS-VNC in either way was less than 1. The agreement on risk stratification with CACS-TNC and CCACS-VNC was almost perfect.

Conclusion

This multicenter study with dual-layer spectral detector CT showed that it was feasible to calculate CACS from the VNC images derived from the spectral coronary artery CT angiography scan, and the results were in good accordance with the TNC images after correction. Therefore, the TNC scan could be omitted, reducing the radiation dose to patients and saving examination time while using dual-layer spectral detector CT.

Accuracy of an Artificial Intelligence Deep Learning Algorithm Implementing a Recurrent Neural Network With Long Short-term Memory for the Automated Detection of Calcified Plaques From Coronary Computed Tomography Angiography

PURPOSE

The purpose of this study was to evaluate the accuracy of a novel fully automated deep learning (DL) algorithm implementing a recurrent neural network (RNN) with long short-term memory (LSTM) for the detection of coronary artery calcium (CAC) from coronary computed tomography angiography (CCTA) data.

MATERIALS AND METHODS

Under an IRB waiver and in HIPAA compliance, a total of 194 patients who had undergone CCTA were retrospectively included. Two observers independently evaluated the image quality and recorded the presence of CAC in the right (RCA), the combination of left main and left anterior descending (LM-LAD), and left circumflex (LCx) coronary arteries. Noncontrast CACS scans were allowed to be used in cases of uncertainty. Heart and coronary artery centerline detection and labeling were automatically performed. Presence of CAC was assessed by a RNN-LSTM. The algorithm's overall and per-vessel sensitivity, specificity, and diagnostic accuracy were calculated.

RESULTS

CAC was absent in 84 and present in 110 patients. As regards CCTA, the median subjective image quality, signal-to-noise ratio, and contrast-to-noise ratio were 3.0, 13.0, and 11.4. A total of 565 vessels were evaluated. On a per-vessel basis, the algorithm achieved a sensitivity, specificity, and diagnostic accuracy of 93.1% (confidence interval [CI], 84.3%-96.7%), 82.76% (CI, 74.6%-89.4%), and 86.7% (CI, 76.8%-87.9%), respectively, for the RCA, 93.1% (CI, 86.4%-97.7%), 95.5% (CI, 88.77%-98.75%), and 94.2% (CI. 90.2%-94.6%), respectively, for the LM-LAD, and 89.9% (CI, 80.2%-95.8%), 90.0% (CI, 83.2%-94.7%), and 89.9% (CI, 85.0%-94.1%), respectively, for the LCx. The overall sensitivity, specificity, and diagnostic accuracy were 92.1% (CI, 92.1%-95.2%), 88.9% (CI. 84.9%-92.1%), and 90.3% (CI, 88.0%-90.0%), respectively. When accounting for image quality, the algorithm achieved a sensitivity, specificity, and diagnostic accuracy of 76.2%, 87.5%, and 82.2%, respectively, for poor-quality data sets and 93.3%, 89.2% and 90.9%, respectively, when data sets rated adequate or higher were combined.

CONCLUSION

The proposed RNN-LSTM demonstrated high diagnostic accuracy for the detection of CAC from CCTA.

Automatic coronary calcium scoring in chest CT using a deep neural network in direct comparison with non-contrast cardiac CT: A validation study

PURPOSE

To evaluate deep-learning based calcium quantification on Chest CT scans compared with manual evaluation, and to enable interpretation in terms of the traditional Agatston score on dedicated Cardiac CT.

METHODS

Automated calcium quantification was performed using a combination of deep-learning convolution neural networks with a ResNet-architecture for image features and a fully connected neural network for spatial coordinate features. Calcifications were identified automatically, after which the algorithm automatically excluded all non-coronary calcifications using coronary probability maps and aortic segmentation. The algorithm was first trained on cardiac-CTs and refined on non-triggered chest-CTs. This study used on 95 patients (cohort 1), who underwent both dedicated calcium scoring and chest-CT acquisitions using the Agatston score as reference standard and 168 patients (cohort 2) who underwent chest-CT only using qualitative expert assessment for external validation. Results from the deep-learning model were compared to Agatston-scores(cardiac-CTs) and manually determined calcium volumes(chest-CTs) and risk classifications.

RESULTS

In cohort 1, the Agatston score and AI determined calcium volume shows high correlation with a correlation coefficient of 0.921(p < 0.001) and R² of 0.91. According to the Agatston categories, a total of 67(70 %) were correctly classified with a sensitivity of 91 % and specificity of 92 % in detecting presence of coronary calcifications. Manual determined calcium volume on chest-CT showed excellent correlation with the AI volumes with a correlation coefficient of 0.923(p < 0.001) and R² of 0.96, no significant difference was found (p = 0.247). According to qualitative risk classifications in cohort 2, 138(82 %) cases were correctly classified with a k-coefficient of 0.74, representing good agreement. All wrongly classified scans (30(18 %)) were attributed to an adjacent category.

CONCLUSION

Artificial intelligence based calcium quantification on chest-CTs shows good correlation compared to reference standards. Fully automating this process may reduce evaluation time and potentially optimize clinical calcium scoring without additional acquisitions.

Coronary artery calcium scoring on non-gated, non-contrast chest computed tomography (CT) using wide-detector, high-pitch and fast gantry rotation: comparison with dedicated calcium scoring CT

Background

Our study assessed the reliability of non-gated, non-contrast chest computed tomography (NCCT) (with high pitch, wide coverage, and fast gantry rotation time, reconstructed at various slice thicknesses), compared with the electrocardiography (ECG)-gated calcium scoring cardiac computed tomography (CaCT), for quantifying coronary artery calcification (CAC).

Methods

Patients aged ≥50 years who required clinical NCCT were prospectively enrolled. All CT scans were performed with 256-detector rows; z-axis coverage, 8 cm; pitch, 1.5; and gantry rotation time, 280 ms (table feed, 42.86 cm/s). NCCT was followed by ECG-gated CaCT. The NCCT images were reconstructed at 0.625-, 1.25-, and 2.5-mm slice intervals. The CAC score was calculated on four sets of CT images with a commercially available software using the Agatston method. The CAC scores were divided into four standard Agatston scoring categories (Agatston scores: 0, 1–100, 101–400, and >400). The inter-observer and inter-technique agreements were evaluated for the CAC scores.

Results

Twenty-six patients (M:F, 14:12; mean age, 66.04±6.97 years) were evaluated. Agatston scores showed near-perfect correlation between CaCT and NCCT for each slice thickness. On Bland-Altman analysis, the mean differences of Agatston scores between CaCT and NCCT (slice thicknesses: 0.625, 1.25, and 2.5 mm) were 37.54, 6.67, and −41.04, respectively. Inter-technique concordance was high for the four Agatston scoring categories with linear-weighted kappa values of 0.599, 0.609, and 0.597 for NCCT (slice thicknesses: 0.625, 1.25, and 2.5 mm, respectively). NCCT with 1.25-mm slice thickness showed the strongest correlation with CaCT.

Conclusions

CAC quantification with NCCT using a wide detector, high pitch, and high temporal resolution scanning modes correlates very highly with ECG-gated CaCT, and 1.25-mm slice thickness NCCT images are more reliable than other NCCT images.

Sinogram-Affirmed Iterative Reconstruction Negatively Impacts the Risk Category Based on Agatston Score: A Study Combining Coronary Calcium Score Measurement and Coronary CT Angiography

Purpose

To assess the impact of sinogram-affirmed iterative reconstruction (SAFIRE) on risk category for coronary artery disease by combining coronary calcium score measurement and coronary CT angiography (CCTA).

Materials and Methods

Eighty-nine patients (64.0% male) older than 18 years (64.4 ± 10.3 years) underwent coronary artery calcium scanning and prospectively ECG-triggered sequential CCTA examination. All raw data acquired in coronary artery calcium scanning were reconstructed by both filtered back projection (FBP) and SAFIRE algorithms with 5 different levels. Objective image quality and calcium quantification were evaluated and compared between FBP and all SAFIRE levels by the Sphericity Assumed test or Greenhouse-Geisser ε correction coefficient. Coronary artery stenosis was assessed in CCTA. Risk categories of all patients and of the patients with coronary artery stenosis in CCTA were compared between FBP and all SAFIRE levels by the Friedman test.

Results

The reconstruction protocol from traditional FBP to SAFIRE 5 was associated with a gradual reduction in CT value and image noise (P < 0.001) but associated with a gradual improvement in the signal-to-noise ratio (P < 0.001). There was a gradual reduction in coronary calcification quantification (Agatston score: from 73.5 in FBP to 38.1 in SAFIRE 5, P < 0.001) from traditional FBP to SAFIRE 5. There was a significant difference for the risk category between FBP and all levels of SAFIRE in all patients (from 3.5 in FBP to 3.2 in SAFIRE 5, P < 0.001) and in the patients with coronary artery stenosis in CCTA (from 4.0 in FBP to 3.6 in SAFIRE 5, P < 0.001).

Conclusions

SAFIRE significantly reduces coronary calcification quantification compared to FBP, resulting in the reduction of risk categories based on the Agatston score. The risk categories of the patients with coronary artery stenosis in CCTA may also decline. Thus, SAFIRE may lead risk categories to underestimate the existence of significant coronary artery stenosis.

Technical feasibility and validation of a coronary artery calcium scoring system using CT coronary angiography images

OBJECTIVES

We validate a novel CT coronary angiography (CCTA) coronary calcium scoring system.

METHODS

Calcium was quantified on CCTA images using a new patient-specific attenuation threshold: mean + 2SD of intra-coronary contrast density (HU). Using 335 patient data sets a conversion factor (CF) for predicting CACS from CCTA scores (CCTAS) was derived and validated in a separate cohort (n = 168). Bland-Altman analysis and weighted kappa for MESA centiles and Agatston risk groupings were calculated.

RESULTS

Multivariable linear regression yielded a CF: CACS = (1.185 × CCTAS) + (0.002 × CCTAS × attenuation threshold). When applied to CCTA data sets there was excellent correlation (r = 0.95; p < 0.0001) and agreement (mean difference -10.4 [95% limits of agreement -258.9 to 238.1]) with traditional calcium scores. Agreement was better for calcium scores below 500; however, MESA percentile agreement was better for high risk patients. Risk stratification was excellent (Agatston groups k = 0.88 and MESA centiles k = 0.91). Eliminating the dedicated CACS scan decreased patient radiation exposure by approximately one-third.

CONCLUSION

CCTA calcium scores can accurately predict CACS using a simple, individualized, semiautomated approach reducing acquisition time and radiation exposure when evaluating patients for CAD. This method is not affected by the ROI location, imaging protocol, or tube voltage strengthening its clinical applicability.

KEY POINTS

• Coronary calcium scores can be reliably determined on contrast-enhanced cardiac CT • This score can accurately risk stratify patients • Elimination of a dedicated calcium scan reduces patient radiation by a third.

Scoring of coronary artery calcium scans: history, assumptions, current limitations, and future directions

Coronary artery calcium (CAC) scanning is a reliable, noninvasive technique for estimating overall coronary plaque burden and for identifying risk for future cardiac events. Arthur Agatston and Warren Janowitz published the first technique for scoring CAC scans in 1990. Given the lack of available data correlating CAC with burden of coronary atherosclerosis at that time, their scoring algorithm was remarkable, but somewhat arbitrary. Since then, a few other scoring techniques have been proposed for the measurement of CAC including the Volume score and Mass score. Yet despite new data, little in this field has changed in the last 15 years. The main focus of our paper is to review the implications of the current approach to scoring CAC scans in terms of correlation with the central disease - coronary atherosclerosis. We first discuss the methodology of each available scoring system, describing how each of these scores make important indirect assumptions in the way they account (or do not account) for calcium density, location of calcium, spatial distribution of calcium, and microcalcification/emerging calcium that might limit their predictive power. These assumptions require further study in well-designed, large event-driven studies. In general, all of these scores are adequate and are highly correlated with each other. Despite its age, the Agatston score remains the most extensively studied and widely accepted technique in both the clinical and research settings. After discussing CAC scoring in the era of contrast enhanced coronary CT angiography, we discuss suggested potential modifications to current CAC scanning protocols with respect to tube voltage, tube current, and slice thickness which may further improve the value of CAC scoring. We close with a focused discussion of the most important future directions in the field of CAC scoring.

Automatic detection and quantification of the Agatston coronary artery calcium score on contrast computed tomography angiography

Potentially, Agatston coronary artery calcium (CAC) score could be calculated on contrast computed tomography coronary angiography (CTA). This will make a separate non-contrast CT scan superfluous. This study aims to assess the performance of a novel fully automatic algorithm to detect and quantify the Agatston CAC score in contrast CTA images. From a clinical registry, 20 patients were randomly selected for each CAC category (i.e. 0, 1-99, 100-399, 400-999, ≥1,000). The Agatston CAC score on non-contrast CT was calculated manually, while the novel algorithm was used to automatically detect and quantify Agatston CAC score in contrast CTA images. The resulting Agatston CAC scores were validated against the non-contrast images. A total of 100 patients (60 ± 11 years, 63 men) were included. The median CAC score on non-contrast CT was 145 (IQR 5-760), whereas the contrast CTA CAC score was 170 (IQR 23-594) (P = 0.004). The automatically computed CAC score showed a high correlation (R = 0.949; P < 0.001) and intra-class correlation (R = 0.863; P < 0.001) with non-contrast CT CAC score. Moreover, agreement within CAC categories was good (κ 0.588). Fully automatic detection of Agatston CAC score on contrast CTA is feasible and showed high correlation with non-contrast CT CAC score. This could imply a radiation dose reduction and time saving by omitting the non-contrast scan.

Automatic assessment of coronary artery calcium score from contrast-enhanced 256-row coronary computed tomography angiography

The coronary artery calcium score (CS), an independent predictor of cardiovascular events, can be obtained from a stand-alone nonenhanced computed tomography (CT) scan (CSCT) or as an additional nonenhanced procedure before contrast-enhanced coronary CT angiography (CCTA). We evaluated the accuracy of a novel fully automatic tool for computing CS from the CCTA examination. One hundred thirty-six consecutive symptomatic patients (aged 59 ± 11 years, 40% female) without known coronary artery disease who underwent both 256-row CSCT and CCTA were studied. Original scan reconstruction (slice thickness) was maintained (3 mm for CSCT and 0.67 mm for CCTA). CS was computed from CCTA by an automatic tool (COR Analyzer, rcadia Medical Imaging, Haifa, Israel) and compared with CS results obtained by standard assessment of nonenhanced CSCT (HeartBeat CS, Philips, Cleveland, Ohio). We also compared both methods for classification into 5 commonly used CS categories (0, 1 to 10, 11 to 100, 101 to 400, >400 Agatston units). All scans were of diagnostic quality. CS obtained by the COR Analyzer from CCTA classified 111 of 136 (82%) of patients into identical categories as CS by CSCT and 24 of remaining 25 into an adjacent category. Overall, CS values from CCTA showed high correlation with CS values from CSCT (Spearman rank correlation = 0.95, p <0.0001). In conclusion, CS values automatically computed from 256-row CCTA correlated highly with standard CS values obtained from nonenhanced CSCT. CS obtained directly from CCTA may obviate the need for an additional scan and attendant radiation.

Fully automatic model-based calcium segmentation and scoring in coronary CT angiography

PURPOSE

The paper presents new methods for automatic coronary calcium detection, segmentation and scoring in coronary CT angiography (cCTA) studies.

METHODS

Calcium detection and segmentation are performed by modeling image intensity profiles of coronary arteries. The scoring algorithm is based on a simulated unenhanced calcium score (CS) CT image, constructed by virtually removing the contrast media from cCTA. The methods are implemented as part of a fully automatic system for CS assessment from cCTA.

RESULTS

The system was tested in two independent clinical trials on 263 studies and demonstrated 0.95/0.91 correlation between the CS computed from cCTA and the standard Agatston score derived from unenhanced CS CT. The mean absolute percent difference (MAPD) of 36/39 % between the two scores lies within the error range of the standard CS CT (15-65 %).

CONCLUSIONS

High diagnostic performance, combined with the benefits of the fully automatic solution, suggests that the proposed technique can be used to eliminate the need in a separate CS CT scan as part of the cCTA examination, thus reducing the radiation exposure and simplifying the procedure.

Automated quantification of epicardial adipose tissue using CT angiography: evaluation of a prototype software

OBJECTIVES

This study evaluated the performance of a novel automated software tool for epicardial fat volume (EFV) quantification compared to a standard manual technique at coronary CT angiography (cCTA).

METHODS

cCTA data sets of 70 patients (58.6 ± 12.9 years, 33 men) were retrospectively analysed using two different post-processing software applications. Observer 1 performed a manual single-plane pericardial border definition and EFVM segmentation (manual approach). Two observers used a software program with fully automated 3D pericardial border definition and EFVA calculation (automated approach). EFV and time required for measuring EFV (including software processing time and manual optimization time) for each method were recorded. Intraobserver and interobserver reliability was assessed on the prototype software measurements. T test, Spearman's rho, and Bland-Altman plots were used for statistical analysis.

RESULTS

The final EFVA (with manual border optimization) was strongly correlated with the manual axial segmentation measurement (60.9 ± 33.2 mL vs. 65.8 ± 37.0 mL, rho = 0.970, P < 0.001). A mean of 3.9 ± 1.9 manual border edits were performed to optimize the automated process. The software prototype required significantly less time to perform the measurements (135.6 ± 24.6 s vs. 314.3 ± 76.3 s, P < 0.001) and showed high reliability (ICC > 0.9).

CONCLUSIONS

Automated EFVA quantification is an accurate and time-saving method for quantification of EFV compared to established manual axial segmentation methods.

KEY POINTS

• Manual epicardial fat volume quantification correlates with risk factors but is time-consuming. • The novel software prototype automates measurement of epicardial fat volume with good accuracy. • This novel approach is less time-consuming and could be incorporated into clinical workflow.