OBJECTIVES

This study was conducted to investigate the influence of coronary artery calcium (CAC) score on the diagnostic performance of machine-learning-based coronary computed tomography (CT) angiography (cCTA)-derived fractional flow reserve (CT-FFR).

BACKGROUND

CT-FFR is used reliably to detect lesion-specific ischemia. Novel CT-FFR algorithms using machine-learning artificial intelligence techniques perform fast and require less complex computational fluid dynamics. Yet, influence of CAC score on diagnostic performance of the machine-learning approach has not been investigated.

METHODS

A total of 482 vessels from 314 patients (age 62.3 +/- 9.3 years, 77% male) who underwent cCTA followed by invasive FFR were investigated from the MACHINE (Machine Learning based CT Angiography derived FFR: a Multi-center Registry) registry data. CAC scores were quantified using the Agatston convention. The diagnostic performance of CT-FFR to detect lesion-specific ischemia was assessed across all Agatston score categories (CAC 0, >0 to <100, 100 to <400, and >=$400) on a per-vessel level with invasive FFR as the reference standard.

RESULTS

The diagnostic accuracy of CT-FFR versus invasive FFR was superior to cCTA alone on a per-vessel level (78% vs. 60%) and per patient level (83% vs. 73%) across all Agatston score categories. No statistically significant differences in the diagnostic accuracy, sensitivity, or specificity of CT-FFR were observed across the categories. CT-FFR showed good discriminatory power in vessels with high Agatston scores (CAC >= 400) and high performance in low-to-intermediate Agatston scores (CAC >0 to <400) with a statistically significant difference in the area under the receiver-operating characteristic curve (AUC) (AUC: 0.71 [95% confidence interval (CI): 0.57 to 0.85] vs. 0.85 [95% CI: 0.82 to 0.89], p = 0.04). CT-FFR showed superior diagnostic value over cCTA in vessels with high Agatston scores (CAC >= 400: AUC 0.71 vs. 0.55, p = 0.04) and low-to-intermediate Agatston scores (CAC >0 to <400: AUC 0.86 vs. 0.63, p < 0.001).

CONCLUSIONS

Machine-learning-based CT-FFR showed superior diagnostic performance over cCTA alone in CAC with a significant difference in the performance of CT-FFR as calcium burden/Agatston calcium score increased. (Machine Learning Based CT Angiography Derived FFR: a Multicenter, Registry [MACHINE] NCT02805621). (C) 2020 by the American College of Cardiology Foundation.

Coronary computed tomography angiography-derived fractional flow reserve (CT-FFR) is a noninvasive application to evaluate the hemodynamic impact of coronary artery disease by simulating invasively measured FFR based on CT data. CT-FFR is based on the assumption of a normal coronary microvascular response. We assessed the diagnostic performance of a machine-learning based application for on-site computation of CT-FFR in patients with and without diabetes mellitus with suspected coronary artery disease. The study population included 75 diabetic and 276 nondiabetic patients who were enrolled in the MACHINE consortium. The overall diagnostic performance of coronary CT angiography alone and in combination with CT-FFR were analyzed with direct invasive FFR comparison in 110 coronary vessels of the diabetic group and in 415 coronary vessels of the nondiabetic group. Per-vessel discrimination of lesion-specific ischemia by CT-FFR was assessed by the area under the receiver operating characteristic curves. The overall diagnostic accuracy of CT-FFR in diabetic patients was 83% and in nondiabetic patients 75% (p = 0.088), showing improvement over the diagnostic accuracy of coronary CT angiography, which was 58% and 65% (p = 0.223), respectively. In addition, the diagnostic accuracy of CT-FFR was similar between diabetic and nondiabetic patients per stratified CT-FFR group (CT-FFR amp;lt; 0.6, 0.6 to 0.69, 0.7 to 0.79, 0.8 to 0.89, amp;gt;= 0.9). The area under the curves for diabetic and nondiabetic patients were also comparable, 0.88 and 0.82 (p = 0.113), respectively. In conclusion, on-site machine-learning CT-FFR analysis improved the diagnostic performance of coronary CT angiography and accurately discriminated lesion-specific ischemia in both diabetic and nondiabetic patients suspected of coronary artery disease. (C) 2018 Elsevier Inc. All rights reserved.

Background: Coronary computed tomographic angiography (CTA) is a reliable modality to detect coronary artery disease. However, CTA generally overestimates stenosis severity compared with invasive angiography, and angiographic stenosis does not necessarily imply hemodynamic relevance when fractional flow reserve (FFR) is used as reference. CTA-based FFR (CT-FFR), using computational fluid dynamics (CFD), improves the correlation with invasive FFR results but is computationally demanding. More recently, a new machine-learning (ML) CT-FFR algorithm has been developed based on a deep learning model, which can be performed on a regular workstation. In this large multicenter cohort, the diagnostic performance ML-based CT-FFR was compared with CTA and CFD-based CT-FFR for detection of functionally obstructive coronary artery disease. Methods and Results: At 5 centers in Europe, Asia, and the United States, 351 patients, including 525 vessels with invasive FFR comparison, were included. ML-based and CFD-based CT-FFR were performed on the CTA data, and diagnostic performance was evaluated using invasive FFR as reference. Correlation between ML-based and CFD-based CT-FFR was excellent (R=0.997). ML-based (area under curve, 0.84) and CFD-based CT-FFR (0.84) outperformed visual CTA (0.69; Pamp;lt;0.0001). On a per-vessel basis, diagnostic accuracy improved from 58% (95% confidence interval, 54%-63%) by CTA to 78% (75%-82%) by ML-based CT-FFR. The per-patient accuracy improved from 71% (66%-76%) by CTA to 85% (81%-89%) by adding ML-based CT-FFR as 62 of 85 (73%) false-positive CTA results could be correctly reclassified by adding ML-based CT-FFR. Conclusions: On-site CT-FFR based on ML improves the performance of CTA by correctly reclassifying hemodynamically nonsignificant stenosis and performs equally well as CFD-based CT-FFR.

Background

Cardiac Computed Tomography Angiography (CCTA) is becoming increasingly useful in the work‐up of coronary artery disease (CAD). Several potential methods for increasing the diagnostic yield of cardiac CT are available.

Purpose

Study I: To investigate whether the use of a 2‐D, non‐linear adaptive noise reduction filter can improve CCTA image quality.

Study II: To evaluate the variation in adenosine stress dynamic CT perfusion (CTP) blood flow as compared to stress 99mTc SPECT. Secondly, to compare the perfusion results from manual and automatic myocardial CTP segmentation.

Study III: To evaluate the accuracy of non‐invasive, CCTA‐derived Fractional Flow Reserve (cFFR).

Study IV: To evaluate the prognostic value of CCTA in terms of major adverse cardiac events (MACE).

Materials and methods

Study I: Single images from 36 consecutive CCTA exams performed with two different dose levels were used. Image quality in full dose, low‐dose and noise‐reduced low‐dose images was graded using visual grading analysis. Image noise was measured.

Study II: CTP and SPECT were performed in 17 patients, and the variation in per AHA‐segment blood flow was evaluated and compared. CTP results from manual and automated image segmentation were compared.

Study III: CCTA datasets from 21 patients were processed using cFFR software and the results compared to the corresponding invasively measured FFR (invFFR).

Study IV: 1205 consecutive patients with chest pain of unknown origin underwent CCTA. Baseline data and data on subsequent MACE were retrieved from relevant registries. Survival, hazard ratios and the three‐year incidence of cardiac events and readmissions were calculated.

Results

Study I: There was significant improvement in perceived image quality for all criteria when the filter was applied, and a significant decrease in image noise.

Study II: The correlation coefficients for CTP vs. SPECT were 0.38 and 0.41 (p<0.001, for manual and automated segmentation respectively. Mean per patient CTP blood flow in normal segments varied between 94‐183 ml/100 ml tissue/min for manual segmentation, and 104‐196 ml/100 ml tissue/min for automated segmentation. The Spearman rank correlation coefficient for manual vs. automated segmentation CTP was ρ = 0.88 (p<0.001) and the Intraclass Correlation Coefficient (ICC) was 0.93 (p<0.001).

Study III: The Spearman rank correlation coefficient for cFFR vs. invFFR was ρ = 0.77 (p<0.001) and the ICC was 0.73 (p<0.001). Sensitivity, specificity, positive predictive value and negative predictive value for significant stenosis (FFR<0.80, per vessel) were 0.83, 0.76, 0.56 and 0.93 respectively.

Study IV: The hazard ratio for non‐obstructive CAD vs. normal coronary arteries was 5.13 (95% C.I 1.03‐25.43, p<0.05), and 151.40 (95% C.I 37.03‐619.08, p<0.001) for obstructive CAD vs. normal coronary arteries. The three‐year incidence of MACE was 1.1% for patients with normal vessels on CCTA, 2.5% for patients with non‐obstructive CAD and 42.7% for patients with obstructive CAD (p<0.001).

Conclusions:

Study I: Image quality and noise levels of low dose images were significantly improved with the filter, even though the improvement was small compared to the image quality of the corresponding diastolic full‐dose images.

Study II: Correlation between dynamic CTP and SPECT was positive but weak. There were large variations in CTP blood flow in normal segments on SPECT, rendering the definition of an absolute cut‐off value for normal vs. ischemic myocardium difficult. Manual and automatic segmentation were equally useful.

Study III: The correlation between cFFR and invFFR was good, indicating that noninvasively estimated cFFR performs on a similar level as invasively measure FFR.

Study IV: The long‐term risk for MACE was very low in patients without obstructive CAD on CCTA, though there seemed to be a substantial increase in the risk for MACE even in patients with non‐obstructive CAD as compared to normal coronary arteries. In addition, even patients with normal coronary arteries or non‐obstructive CAD continued to have a substantial number of readmissions for chest pain or angina pectoris.

BACKGROUND: The significance of a coronary stenosis can be determined by measuring the fractional flow reserve (FFR) during invasive coronary angiography. Recently, methods have been developed which claim to be able to estimate FFR using image data from standard coronary computed tomography angiography (CCTA) exams.

PURPOSE: To evaluate the accuracy of non-invasively computed fractional flow reserve (cFFR) from CCTA.

MATERIAL AND METHODS: A total of 23 vessels in 21 patients who had undergone both CCTA and invasive angiography with FFR measurement were evaluated using a cFFR software prototype. The cFFR results were compared to the invasively obtained FFR values. Correlation was calculated using Spearman's rank correlation, and agreement using intraclass correlation coefficient (ICC). Sensitivity, specificity, accuracy, negative predictive value, and positive predictive value for significant stenosis (defined as both FFR ≤0.80 and FFR ≤0.75) were calculated.

RESULTS: The mean cFFR value for the whole group was 0.81 and the corresponding mean invFFR value was 0.84. The cFFR sensitivity for significant stenosis (FFR ≤0.80/0.75) on a per-lesion basis was 0.83/0.80, specificity was 0.76/0.89, and accuracy 0.78/0.87. The positive predictive value was 0.56/0.67 and the negative predictive value was 0.93/0.94. The Spearman rank correlation coefficient was ρ = 0.77 (P < 0.001) and ICC = 0.73 (P < 0.001).

CONCLUSION: This particular CCTA-based cFFR software prototype allows for a rapid, non-invasive on-site evaluation of cFFR. The results are encouraging and cFFR may in the future be of help in the triage to invasive coronary angiography.

BACKGROUND: Dynamic cardiac CT perfusion (CTP) is based on repeated imaging during the first-pass contrast agent inflow. It is a relatively new method that still needs validation.

PURPOSE: To evaluate the variation in adenosine stress dynamic CTP blood flow as compared to (99m) Tc SPECT. Secondarily, to compare manual and automatic segmentation.

METHODS: Seventeen patients with manifest coronary artery disease were included. Nine were excluded from evaluation for various reasons. All patients were examined with dynamic stress CTP and stress/rest SPECT. CTP blood flow was compared with SPECT on a per segment basis. Results for manual and automated AHA segmentation were compared.

RESULTS: CTP showed a positive correlation with SPECT, with correlation coefficients of 0·38 and 0·41 for manual and automatic segmentation, respectively (P<0·0001). There was no significant difference between the correlation coefficients of the manual and automated segmentation procedures (P = 0·75). The average per individual global CTP blood flow value for normal segments varied by a factor of 1·9 (manual and automatic segmentation). For the whole patient group, the CTP blood flow value in normal segments varied by a factor of 2·9/2·7 (manual/automatic segmentation). Within each patient, the average per segment blood flow in normal segments varied by a factor of 1·3-2·0/1·2-2·1 (manual/automatic segmentation).

CONCLUSION: A positive but rather weak correlation was found between CTP and (99m) Tc SPECT. Large variations in CTP blood flow suggest that a cut-off value for stress myocardial blood flow is inadequate to detect ischaemic segments. Dynamic CTP is hampered by a limited coverage.

Objectives To propose a method to study the effect of exposure settings on image quality and to estimate the potential for dose reduction when introducing dose-reducing measures.

Methods Using the framework of visual grading regression (VGR), a log(mAs) term is included in the ordinal logistic regression equation, so that the effect of reducing the dose can be quantitatively related to the effect of adding post-processing. In the ordinal logistic regression, patient and observer identity are treated as random effects using generalised linear latent and mixed models. The potential dose reduction is then estimated from the regression coefficients. The method was applied in a single-image study of coronary CT angiography (CTA) to evaluate two-dimensional (2D) adaptive filters, and in an image-pair study of abdominal CT to evaluate 2D and three-dimensional (3D) adaptive filters.

Results For five image quality criteria in coronary CTA, dose reductions of 16–26% were predicted when adding 2D filtering. Using five image quality criteria for abdominal CT, it was estimated that 2D filtering permits doses were reduced by 32–41%, and 3D filtering by 42–51%.

Conclusions VGR including a log(mAs) term can be used for predictions of potential dose reduction that may be useful for guiding researchers in designing subsequent studies evaluating diagnostic value. With appropriate statistical analysis, it is possible to obtain direct numerical estimates of the dose-reducing potential of novel acquisition, reconstruction or post-processing techniques.

Purpose: To retrospectively evaluate the diagnostic accuracy of coronary CT angiography (CCTA) using segmented 3D data for the detection of significant stenoses with catheter angiography (CA) as the reference standard.

Method: CCTA data sets from 30 patients were acquired with a 64-slice dual source CT scanner and segmented by an independent observer using the region growing (RG) method and the “virtual contrast injection” (VC) method. For every examination, each of the three types of images was  then reviewed by one of three reviewers in a blinded fashion for the presence of stenoses with diameter reduction of 50% or more. For the original series, the reviewer was allowed to use all the 2D or 3D visualization tools available (mixed method). For the segmented results (from RG and VC), the reviewer only used the 3D maximum intensity projection. Evaluation results were compared with CA for each artery.

Results: Overall, 34 arteries with significant stenosis were identified by CA. The percentage of evaluable arteries, accuracy and negative predictive value (NPV) for detecting stenosis were, respectively, 86%, 74% and 93% for the mixed method, 83%, 71% and 92% for VC, and 64%, 56% and 93% for RG. Accuracy was significantly lower for the RG method than for the other two methods (p<0.01), whereas there was no significant difference in accuracy between the VC method and the mixed method (p = 0.22). Excluding vessels with heavy calcification, all three methods had similar accuracy.

Conclusion: Diagnostic accuracy when using segmented 3D data was lower than with access to 2D images. However, the high NPV of the 3D methods suggests a potential of using them as an initial step, with access to 2D reviewing techniques for suspected lesions and cases with heavy calcification. The VC method, which generates more evaluable arteries and has higher accuracy, seems more promising for this purpose than the RG method.

For optimization and evaluation of image quality, one can use visual grading experiments, where observers rate some aspect of image quality on an ordinal scale. To take into account the ordinal character of the data, ordinal logistic regression is used in the statistical analysis, an approach known as visual grading regression (VGR). In the VGR model one may include factors such as imaging parameters and post-processing procedures, in addition to patient and observer identity. In a single-image study, 9 radiologists graded 24 cardiac CTA images acquired with ECG-modulated tube current using standard settings (310 mAs), reduced dose (62 mAs) and reduced dose after post-processing. Image quality was assessed using visual grading with five criteria, each with a five-level ordinal scale from 1 (best) to 5 (worst). The VGR model included one term estimating the dose effect (log of mAs setting) and one term estimating the effect of postprocessing. The model predicted that 115 mAs would be required to reach an 80% probability of a score of 1 or 2 for visually sharp reproduction of the heart without the post-processing filter. With the post-processing filter, the corresponding figure would be 86 mAs. Thus, applying the post-processing corresponded to a dose reduction of 25%. For other criteria, the dose-reduction was estimated to 16-26%. Using VGR, it is thus possible to quantify the potential for dose-reduction of post-processing filters.

To analyze visual grading experiments, ordinal logistic regression (here called visual grading regression, VGR) may be used in the statistical analysis. In addition to types of imaging or post-processing, the VGR model may include factors such as patient and observer identity, which should be treated as random effects. Standard software does not allow random factors in ordinal logistic regression, but using Generalized Linear Latent And Mixed Models (GLLAMM) this is possible. In a single-image study, 9 radiologists graded 24 cardiac Computed Tomography Angiography (CTA) images with reduced dose without and after post-processing with a 2D adaptive filter, using five image quality criteria. First, standard ordinal logistic regression was carried out, treating filtering, patient and observer identity as fixed effects. The same analysis was then repeated with GLLAMM, treating filtering as a fixed effect and patient and observer identity as random effects. With both approaches, a significant effect (pless than0.01) of the filtering was found for all five criteria. No dramatic differences in parameter estimates or significance levels were found between the two approaches. It is concluded that random effects can be appropriately handled in VGR using GLLAMM, but no major differences in the results were found in a preliminary evaluation.