First comparison between artificial intelligence-guided coronary computed tomography angiography versus single-photon emission computed tomography testing for ischemia in clinical practice

Abstract

Background

Noninvasive cardiac testing with coronary computed tomography angiography (CCTA) and single-photon emission computed tomography (SPECT) are becoming alternatives to invasive angiography for the evaluation of obstructive coronary artery disease. We aimed to evaluate whether a novel artificial intelligence (AI)-assisted CCTA program is comparable to SPECT imaging for ischemic testing.

Methods

CCTA images were analyzed using an artificial intelligence convolutional neural network machine-learning-based model, atherosclerosis imaging-quantitative computed tomography (AI-QCT)_ISCHEMIA. A total of 183 patients (75 females and 108 males, with an average age of 60.8 years ± 12.3 years) were selected. All patients underwent AI-QCT_ISCHEMIA-augmented CCTA, with 60 undergoing concurrent SPECT and 16 having invasive coronary angiograms. Eight studies were excluded from analysis due to incomplete data or coronary anomalies.

Results

 A total of 175 patients (95%) had CCTA performed, deemed acceptable for AI-QCT_ISCHEMIA interpretation. Compared to invasive angiography, AI-QCT_ISCHEMIA-driven CCTA showed a sensitivity of 75% and specificity of 70% for predicting coronary ischemia, versus 70% and 53%, respectively for SPECT. The negative predictive value was high for female patients when using AI-QCT_ISCHEMIA compared to SPECT (91% vs. 68%, P = 0.042). Area under the receiver operating characteristic curves were similar between both modalities (0.81 for AI-CCTA, 0.75 for SPECT, P = 0.526). When comparing both modalities, the correlation coefficient was r = 0.71 (P < 0.04).

Conclusion

AI-powered CCTA is a viable alternative to SPECT for detecting myocardial ischemia in patients with low- to intermediate-risk coronary artery disease, with significant positive and negative correlation in results. For patients who underwent confirmatory invasive angiography, the results of AI-CCTA and SPECT imaging were comparable. Future research focusing on prospective studies involving larger and more diverse patient populations is warranted to further investigate the benefits offered by AI-driven CCTA.

Introduction

In the USA, cardiovascular disease remains the leading cause of death, with coronary artery disease (CAD) being the leading driver of morbidity and mortality, accounting for nearly 700,000 deaths in 2022 [1]. While invasive coronary angiography remains the gold standard for evaluating obstructive coronary disease, noninvasive tests like coronary computed tomography angiography (CCTA) and single-photon emission computed tomography (SPECT) are continuously evolving technologically, providing safer, more comfortable options for patients. These advancements not only help visualize coronary anatomy but also aid in the early detection and management of CAD, potentially reducing cardiovascular events.

Guidelines from organizations such as the European Society of Cardiology (2019) [2] have indicated that noninvasive testing should be used for patients with suspicion of angina or cardiac etiology causing their symptoms [3]. In actual practice, it is often utilized for screening low- to intermediate-risk patients for CAD [4]. SPECT testing is frequently performed for detecting myocardial ischemia in patients who may have CAD [4]. However, SPECT has drawbacks, such as lower image resolution, the requirement for radioactive tracers, and importantly, the fact that it does not provide actual direct imaging of the coronary anatomy [5].

CCTA is a noninvasive CT imaging modality of cardiac structures and vasculature becoming comparable to invasive coronary angiography for evaluating CAD. The use of AI in CCTA [artificial intelligence (AI)-CCTA] has helped innovate and improve the capacity to risk stratify patients when evaluating the need for a more invasive approach, by increasing image acquisition efficiency and morphology characterization [6].

Despite these advancements, we still need comprehensive, real-world comparisons between AI-enhanced CCTA and traditional SPECT testing. This study aims to assess whether a novel AI-assisted CCTA program [atherosclerosis imaging-quantitative computed tomography (AI-QCT)_ISCHEMIA] is comparable to nuclear medicine stress testing (SPECT) to evaluate for ischemia, similar to noninvasive cardiac testing in real-world practice.

Methods

CCTA images were analyzed via the AI-QCT algorithm, including AI-QCT_ISCHEMIA [7–9] (Cleerly, Inc). The AI-QCT algorithm is commercially available FDA-cleared software that is driven by convolutional neural networks, 3D U-Net, and VGG network variants, to create quantitative output stenosis parameters (e.g. % diameter and area stenosis, the severity of stenosis classification), atherosclerosis parameters (e.g. total plaque volume, noncalcified plaque volume, and lesion size, etc), vessel morphology (total vessel volume, lumen volume, and vessel length, etc), and diffuseness of segments. The AI-QCT results then act as input in the AI-QCT_ISCHEMIA algorithm.

AI-QCT_ISCHEMIA helps determine the risk of myocardial ischemia based on CCTA data by utilizing a random forest plot machine-learning algorithm. AI-QCT_ISCHEMIA determines the probability of abnormal invasive fractional flow reserve (FFR), the gold standard unit of invasive vessel ischemia analysis, via 38 CCTA-derived quantitative variables from Cleerly, Inc. The output is binary, either normal or abnormal.

The AI-QCT_ISCHEMIA performance was compared to invasive coronary angiography vessel-based ischemia defined by either reduced invasive FFR (≤0.8) or instantaneous wave-free ratio (iFR ≤ 0.89) as reference standards. Analysis was restricted to vessels with an invasive measurement available in the main epicardial coronary vessels (left main, anterior descending coronary artery, right coronary artery, and circumflex coronary artery). The performance of AI-QCT_ISCHEMIA was compared to the fractional flow reserve derived from CCTA (Heartflow Inc.).

The AI-QCT_ISCHEMIA algorithm has been validated both internally and externally through multicenter trials, including the CREDENCE (computed tomographic evaluation of atherosclerotic determinants of myocardial ischemia), CLARIFY-2 (AI evaluation of stenosis on CCTA, comparison with quantitative coronary angiography and FFR), and PACIFIC-1 (prospective comparison of cardiac PET/CT, SPECT/CT perfusion imaging and CT coronary angiography with invasive coronary angiography) studies [10–12]. These trials evaluated the effectiveness of AI-QCT_ISCHEMIA compared to coronary angiography as the gold standard in predicting ischemia and CAD and found the two methods to be comparatively accurate.

The performance of AI-QCT_ISCHEMIA as compared to coronary angiography was evaluated by comparing sensitivity, specificity, and both positive predictive values and negative predictive values (NPVs) between AI-CCTA and SPECT testing. Furthermore, an area under the receiver operating characteristic curve (AUC) analysis was used to evaluate diagnostic accuracy between different modalities using the DeLong test for assessment of statistical significance. Statistical analyses were conducted using GraphPad and Microsoft Excel.

Results

This study included 183 patients (75 females and 108 males, with an average age of 60.8 years ± 12.3 years) who were deemed at intermediate risk for CAD based on atypical chest pain. All patients underwent AI-QCT_ISCHEMIA-driven CCTA, with 60 also undergoing concurrent SPECT and 16 having invasive coronary angiograms. Though coronary angiography was used as the gold standard for comparison, it was not used in all patients due to its invasive nature. Importantly, patients who tested negative for ischemia on SPECT or CCTA, or who refused catheterization, were not compared to that gold standard to mitigate unnecessary risk. The entire subset of 16 patients who underwent coronary angiography was incorporated into the sensitivity and specificity. The additional patients incorporated into the analysis for ischemia had tested positive for ischemia with SPECT or CCTA. Eight studies were excluded from analysis due to incomplete data or coronary anomalies.

A total of 175 patients (95%) had CCTA performed, which was deemed acceptable for AI-QCT_ISCHEMIA interpretation. Compared to invasive angiography, AI-QCT_ISCHEMIA-driven CCTA showed a sensitivity of 75% and specificity of 70% for predicting coronary ischemia. For SPECT, values were 70% and 53%, respectively (Table 1).

Table 1.

Coronary computed tomography angiography (AI-QCT_ISCHEMIA) versus single-photon emission computed tomography for coronary ischemia evaluation

Endpoint	AI-QCTISCHEMIA	SPECT	P value
Sensitivity	75% (14/17)	70% (16/17)	0.048
Specificity	70% (19/25)	53% (11/25)	0.039
NPV*	91% (33/42)	68% (27/42)	0.042
AUC	0.81	0.76	0.526

AI-QCT, atherosclerosis imaging-quantitative computed tomography; AUC, area under the curve; NPV, negative predictive value; SPECT, single-photon emission computed tomography.

^*Female patients.

Additionally, the NPV was particularly high for female patients when using AI-QCT_ISCHEMIA, with an NPV of 91% compared to 68% for other methods (P = 0.042) (Table 1), highlighting a potential gender-specific advantage of AI-augmented CCTA in reducing unnecessary interventions in female patients. AUCs were similar between the two modalities, with AI-QCT_ISCHEMIA showing an AUC of 0.81 and SPECT showing an AUC of 0.76 (P = 0.526). When comparing the results of AI-QCT_ISCHEMIA-driven CCTA to SPECT directly, the correlation coefficient was r = 0.71 (P < 0.04).

Discussion

Our study demonstrates that the AI-QCT_ISCHEMIA algorithm applied to CCTA is a viable alternative to SPECT for detecting myocardial ischemia in patients with a low to intermediate risk of CAD. When comparing AI-CCTA to SPECT imaging, we demonstrated a strong correlation in both positive and negative directions, as well as similar sensitivity and specificity in ischemia identification. This reinforces the capacity of AI-augmented CCTA to correctly identify cardiac ischemia compared to standard SPECT testing and interpretation. Furthermore, it suggests that AI-guided CCTA imaging can provide results that can ultimately improve our use of traditional noninvasive stress testing modalities.

Additionally, the study shows that the AI-QCT_ISCHEMIA algorithm provides a comprehensive assessment by integrating 38 different parameters and anatomical characteristics such as plaque type, severity of stenosis, and vascular morphology. This offers a 3-in-1 diagnostic approach, extending beyond traditional assessments that only focus on the presence of atherosclerosis and stenosis. Such an approach can help better understand the nature of disease burden and eventually plan for better interventions when proceeding with invasive angiography. A closer look at how our studies allow us to characterize CAD as seen on a coronary angiogram can help us understand the nuances between these modalities (Fig. 1). In Figures 2 and 3, we can see the 3-in-1 diagnostic approach to CAD demonstrated by AI-QCT_ISCHEMIA to more accurately predict ischemia not only just with FFR, as is used by Heartflow, but also by incorporating additional unique analysis points (e.g. stenosis extent, plaque characteristics and morphology, subtypes, and coronary vessels likely to exhibit ischemia) into its interpretation. In comparison, SPECT imaging of the same patient (Fig. 4) shows the potential areas of perfusion defect but lacks the appropriate resolution to actually define the extent of stenosis. It has been previously demonstrated by Chiou et al., [13] that Cleerly AI-CCTA has better specificity, positive predictive values, AUD, and accuracy for ischemic detection when directly compared with Heartflow CCTA and standard CCTA-FFR. Furthermore, in the CREDENCE trial and the CERTAIN substudy, AI-augmented CCTA improved diagnostic accuracy, speed of clinical workflow, and the need for additional downstream testing [11].

Fig. 1.

Invasive left heart coronary angiogram catheterization showing successful PCI of 99% left main into mid-LAD stenosis using overlapping stents (3.5 × 28 mm, 2.75 × 28 mm, 2.5 × 12 mm), with residual moderate disease in the circumflex and mild disease in the RCA. LAD, left anterior descending; LAO, left anterior oblique; PCI, percutaneous coronary intervention, RAO, right anterior oblique; RCA, right coronary artery.

Fig. 2.

Using AI-driven CCTA analysis (Cleerly, Inc.) of the patient’s coronary artery disease, the algorithm was able to determine the degree of stenosis, plaque characteristics, and coronary vessels likely to exhibit ischemia. AI, artificial intelligence; CCTA, coronary computed tomography angiography; LAD, left anterior descending; LM+LAD, left main + left anterior descending.

Fig. 3.

AI-QCT_ISCHEMIA culprit vessel analysis via AI-augmented CCTA of the left main and left anterior descending coronary arteries demonstrating likely ischemia in these territories as well as the plaque composition (total plaque volume, types of plaque present), remodeling index, and diameter/degree of stenosis. AI-QCT, atherosclerosis imaging-quantitative computed tomography; CCTA, coronary computed tomography angiography.

Fig. 4.

Regadenoson myocardial perfusion imaging cardiac stress test of the same patient showing multiple stress-induced perfusion defects in the inferior, anterior, septal, and apical territories. Transient ischemic dilation (TID) of the left ventricle is also noted which is a marker of severe and extensive coronary artery disease.

SPECT imaging remains a viable option when investigating ischemia with higher clinical suspicion for CAD, such as with typical anginal chest pain, and is suitable in these scenarios due to its established accuracy in high-risk populations. Interestingly, however, our study did demonstrate higher NPV when assessing female patients, suggesting that AI-CCTA may help better rule out cardiovascular disease in women’s health, though further studies to evaluate this should be performed in the future.

Ongoing AI cardiac imaging research endeavors are focusing on prospective studies involving larger and more diverse patient populations. While our current study does demonstrate alignment between AI-CCTA and SPECT imaging, we recognize that a more comprehensive database should be reviewed to increase the power of our findings. As discussed above, our study also suggested AI-CCTA is able to rule out cardiovascular disease better in women, therefore further studies looking at the potential gender-specific advantages of CCTA AI should be performed on a larger scale as well. Future directions should also investigate the cost-effectiveness of AI-QCT_ISCHEMIA in routine clinical practice and its role in reducing unnecessary angiographies and improving patient satisfaction. AI-CCTA represents an opportunity to leverage our data-driven society to better evaluate the risks and benefits of an invasive approach to care. Additionally, while this study focused on evaluating patients with low- to intermediate-risk CAD, future studies (and several ongoing trials such as CONFIRM-2, TRANSFORM, etc) will also investigate incorporating such multimodal approaches to care in higher-risk cardiac patients. These studies are also analyzing AI-augmented CCTA’s advantageous ability for individualized patient prognostication of cardiovascular outcomes. This tool therefore has the potential to provide further information not available to help leverage current diagnostics to analyze higher CAD burden.

In conclusion, AI-QCT_ISCHEMIA-driven CCTA presents a promising, efficient alternative to SPECT testing for myocardial ischemia assessment, particularly in low- to intermediate-risk patients. However, larger controlled studies are necessary to validate these findings and explore the full potential of integrating AI into routine CAD evaluation.

Acknowledgements

Conflicts of interest

There are no conflicts of interest.