Explainable Deep Learning Improves Physician Interpretation of Myocardial Perfusion Imaging

Artificial intelligence may improve accuracy of myocardial perfusion imaging (MPI) but will likely be implemented as an aid to physician interpretation rather than an autonomous tool. Deep learning (DL) has high standalone diagnostic accuracy for obstructive coronary artery disease (CAD), but its influence on physician interpretation is unknown. We assessed whether access to explainable DL predictions improves physician interpretation of MPI. Methods: We selected a representative cohort of patients who underwent MPI with reference invasive coronary angiography. Obstructive CAD, defined as stenosis ≥50% in the left main artery or ≥70% in other coronary segments, was present in half of the patients. We used an explainable DL model (CAD-DL), which was previously developed in a separate population from different sites. Three physicians interpreted studies first with clinical history, stress, and quantitative perfusion, then with all the data plus the DL results. Diagnostic accuracy was assessed using area under the receiver-operating-characteristic curve (AUC). Results: In total, 240 patients with a median age of 65 y (interquartile range 58-73) were included. The diagnostic accuracy of physician interpretation with CAD-DL (AUC 0.779) was significantly higher than that of physician interpretation without CAD-DL (AUC 0.747, P = 0.003) and stress total perfusion deficit (AUC 0.718, P < 0.001). With matched specificity, CAD-DL had higher sensitivity when operating autonomously compared with readers without DL results (P < 0.001), but not compared with readers interpreting with DL results (P = 0.122). All readers had numerically higher accuracy with CAD-DL, with AUC improvement 0.02-0.05, and interpretation with DL resulted in overall net reclassification improvement of 17.2% (95% CI 9.2%-24.4%, P < 0.001). Conclusion: Explainable DL predictions lead to meaningful improvements in physician interpretation; however, the improvement varied across the readers, reflecting the acceptance of this new technology. This technique could be implemented as an aid to physician diagnosis, improving the diagnostic accuracy of MPI.

Discordance Between Coronary Artery Calcium Area and Density Predicts Long-Term Atherosclerotic Cardiovascular Disease Risk

BACKGROUND

Coronary artery calcium (CAC) is commonly quantified as the product of 2 generally correlated measures: plaque area and calcium density.

OBJECTIVES

The authors sought to determine whether discordance between calcium area and density has long-term prognostic importance in atherosclerotic cardiovascular disease (ASCVD) risk.

METHODS

The authors studied 10,373 primary prevention participants from the CAC Consortium with CAC >0. Based on their median values, calcium area and mean calcium density were divided into 4 mutually exclusive concordant/discordant groups. Cox proportional hazards regression assessed the association of calcium area/density groups with ASCVD mortality over a median of 11.7 years, adjusting for traditional risk factors and the Agatston CAC score.

RESULTS

The mean age was 56.7 years, and 24% were female. The prevalence of plaque discordance was 19% (9% low calcium area/high calcium density, 10% high calcium area/low calcium density). Female sex (odds ratio [OR]: 1.48 [95% CI: 1.27-1.74]) and body mass index (OR: 0.81 [95% CI: 0.76-0.87], per 5 kg/m2 higher) were significantly associated with high calcium density discordance, whereas diabetes (OR: 2.23 [95% CI: 1.85-3.19]) was most strongly associated with discordantly low calcium density. Compared to those with low calcium area/low calcium density, individuals with low calcium area/high calcium density had a 71% lower risk of ASCVD death (HR: 0.29 [95% CI: 0.09-0.95]).

CONCLUSIONS

For a given CAC score, high calcium density relative to plaque area confers lower long-term ASCVD risk, likely serving as an imaging marker of biological resilience for lesion vulnerability. Additional research is needed to define a robust definition of calcium area/density discordance for routine clinical risk prediction.

Cardiovascular and hepatic disease associations by magnetic resonance imaging: A retrospective cohort study

Background

Hepatic disease is linked to cardiovascular events but the independent association between hepatic and cardiovascular disease remains unclear, given shared risk factors.

Methods

This was a retrospective study of consecutive patients with a clinical cardiac MRI (CMR) and a serological marker of hepatic fibrosis, the FIB-4 score, within one year of clinical imaging. We assessed the relations between FIB-4 scores grouped based on prior literature: low (< 1.3), moderate (1.3–3.25), and high (>3.25), and abnormalities detected by comprehensive CMR grouped into 4 domains: cardiac structure (end diastolic volumes, atrial dimensions, wall thickness); cardiac function (ejection fractions, wall motion abnormalities, cardiac output); vascular structure (ascending aortic and pulmonary arterial sizes); and cardiac composition (late gadolinium enhancement, T1 and T2 times). We used Poisson regression to examine the association between the conventionally defined FIB-4 category (low <1.3, moderate 1.3–3.25, and high >3.25) and any CMR abnormality while adjusting for demographics and traditional cardiovascular risk factors.

Results

Of the 1668 patients studied (mean age: 55.971 ± 7.28, 901 [54%] male), 85.9% had ≥1 cardiac abnormality with increasing prevalence seen within the low (82.0%) to moderate (88.8%) to high (92.3%) FIB-4 categories. Multivariable analyses demonstrated the presence of any cardiac abnormality was significantly associated with having a high-range FIB-4 (prevalence ratio 1.07, 95% CI: 1.01–1.13); notably, the presence of functional cardiac abnormalities were associated with being in the high FIB-4 range (1.41, 1.21–1.65) and any vascular abnormalities with being in the moderate FIB-4 range (1.22, 1.01–1.47).

Conclusions

Elevated FIB-4 was associated with cardiac functional and vascular abnormalities even after adjustment for shared risk factors in a cohort of patients with clinically referred CMR. These CMR findings indicate that cardiovascular abnormalities exist in the presence of subclinical hepatic fibrosis, irrespective of shared risk factors, underscoring the need for further studies of the heart-liver axis.

Cardiometabolic predictors of quantitative high-risk plaque features in a diverse patient population

Background/Introduction

Little is known about the prevalence of high-risk plaque features or cardiometabolic predictors in diverse patient populations with underrepresented minorities, in the setting of stable chest pain.

Purpose

The goals of our study are to 1) describe plaque characteristics in a diverse patient population with underrepresented minorities and 2) characterize cardiometabolic risk factors associated with high prevalence of high-risk quantitative low attenuation noncalcified plaque (LDNCP) burden.

Methods

Our study included patients with chest pain undergoing CCTA between June 2016 and October 2021 for stable chest pain, who had a complete cardiometabolic panel including lipoprotein(a) and lipid panel, and at least one blood pressure recording before CCTA. Patients with prior PCI or CABG where excluded. CACS was performed before CCTA as per Agatston method and quantified in Agatston Units (AU). Stenosis was graded as per SCCT guidelines by cardiologists and radiologists with level 3 cardiac CT expertise. Plaque measurements were performed using previously validated semiautomated software (AutoPlaque version 2.5) in all patients with CAD-RADS &gt;0 by expert readers blinded from patients' characteristics. Coronary atherosclerotic plaque volumes were measured. Independent predictors for plaque on CCTA among patients were examined using Wilcox multivariate logistic regression.

Results

A total of 227 consecutive patients were included in our study (see table; age 55.00 [47.50–62.00] years, 63% female, 16% diabetes, 44% hypertension, 40% hyperlipidemia and 32% with current or previous smoking history). Majority of patients were Hispanic (64%) and the rest were Black (27%), White (6%) and Asian (3%).

Patients with LDNCP burden &gt;4% were older (60.00 [52.00–66.50] vs 53.00 [43.75–61.00]; p&lt;0.001), more likely to be diabetic (27.7 vs 11.5%; p=0.005), hypertensive (67.7 vs 33.8%; p&lt;0.001), hyperlipidemic (64.6 vs 29.9%; p&lt;0.001) and present smokers (31.3 vs 13.9%; p=0.003). Almost all patients (63/67) with LDNCP burden &gt;4% had non-obstructive disease (CAD-RADS&lt;4).

Patient with LDNCP burden &gt;4% were more likely to be on statin therapy (46.0 vs 30.4%; p=0.041). There was no differences in ethnicity, hemoglobin A1C, TC, LDL-C, HLD-C, TGs, lipoprotein(a), SBP or DBP.

By logistic regression analysis, age (OR [CI]: 1.06 [1.01–1.08]), hypertension (2.20, [1.06–4.63]) and hyperlipidemia (2.73 [1.37–5.47]) increased the likelihood of LDNCP burden &gt;4%, but not Lipoprotein (a)&gt;175 nmol/L (OR [CI]: 1.07 [0.48–2.31].

Conclusions

In our cohort of patients with high number of unrepresented minorities presenting with stable chest pain, almost all patients (94%) with LDNCP burden &gt;4% had non-obstructive CAD (CAD-RADS&lt;4). There were no differences in prevalence of LDNCP or CAD-RADS among different ethnic groups. Age, hypertension and hyperlipidemia, were the cardiometabolic factors related to LDNCP burden &gt;4%.

Funding Acknowledgement

Type of funding sources: None.

18F-sodium fluoride positron emission tomography and coronary plaque radiomics derived from computed tomography angiography for prediction of myocardial infarction

Background

Assessments of coronary disease activity with 18F-sodium fluoride positron emission tomography (18F-NaF PET) and radiomics-based precision coronary plaque phenotyping derived from contrast-enhanced computed tomography (CT) have both been shown to enhance risk stratification in patients with coronary artery disease (CAD). To date, no study has investigated whether these two promising methods (which can be obtained during a single imaging session on a hybrid PET/CT scanner) are interchangeable or can provide superior predictive performance when used in combination.

Purpose

We sought to investigate whether the prognostic information provided by latent morphological radiomic coronary plaque features and assessments of disease activity by 18F-NaF PET are complementary in prediction of myocardial infarction.

Methods

Patients with known CAD underwent coronary 18F-NaF PET and CT angiography on a hybrid PET/CT scanner. Coronary 18F-NaF uptake was determined by the coronary microcalcification activity (CMA). We performed quantitative plaque analysis of coronary CT angiography datasets. Additionally, coronary plaque segmentations on CT angiography were used to extract 1103 radiomic features. Using weighted correlation network analysis we derived latent morphological features of coronary plaques which were aggregated to patient-level radiomic normograms to predict myocardial infarction using univariate and multivariate Cox proportional hazard models.

Results

The study cohort comprised of 260 patients with established CAD (age: 65±9 years; 84% men); 179 (69%) participants showed increased coronary 18F-NaF activity (CMA &gt;0). Over 53 [40–59] months of follow-up 18 patients had a myocardial infarction. Using weighted correlation network analysis, from the 1103 radiomic features we derived 15 distinct eigen radiomic features representing latent morphological coronary plaque patterns. On univariate cox modelling 7 of these emerged as predictors of myocardial infarction (Figure). Following adjustments for calcified, noncalcified and low-density noncalcified plaque volumes and 18F-NaF CMA 4 radiomic features (related to texture and geometry) remained independent predictors of myocardial infarction (Figure).

Conclusion(s)

In patients with established CAD latent morphological features of coronary plaques are predictors of myocardial infarction above and beyond plaque volumes and 18F-NaF uptake. Comprehensive plaque analysis with radiomics may enhance risk stratification of CAD patients.

Funding Acknowledgement

Type of funding sources: Public grant(s) – National budget only. Main funding source(s): NIH, Wellcome Trust

Characterizing left ventricular stiffness in women with signs and symptoms of ischemia with no obstructive coronary arteries

Background

Women with signs and symptoms of ischemia and no obstructive coronary arteries (INOCA) have evidence of diastolic dysfunction and are at increased risk of developing heart failure with preserved ejection fraction (HFpEF). However, mechanisms contributing to HFpEF development are poorly understood and often attributed to underlying cardiovascular risk factors.

Purpose

To compare clinical, invasive, and imaging parameters in women with suspected INOCA and various degrees of left ventricular (LV) stiffness (as measured by invasive end-diastolic pressure [EDP]/end diastolic volume [EDV]).

Methods

Women with suspected INOCA underwent invasive LV pressure-volume loop analysis at rest and coronary function testing with a Doppler wire in the left anterior descending artery. Intracoronary vasoactive substances (adenosine, acetylcholine, nitroglycerin) were infused into the left main artery, as published. Rest and adenosine stress cardiac magnetic resonance (CMR) imaging was performed to evaluate LV function, structure, perfusion, and fibrosis. Women in different tertiles of EDP/EDV ratio were compared using t-tests.

Results

A total of 62 women with complete invasive data were included; 2 did not complete CMR. Compared to the lower EDP/EDV tertile, women in the upper tertile were older, had higher ejection fraction, higher mass/volume ratio, worse diastolic function, greater aortic stiffness and worse coronary microvascular function (Table 1). Traditional cardiovascular risk factors were not significantly different.

Conclusion

Among women with INOCA, older age, coronary microvascular dysfunction, and aortic stiffness were related to greater LV stiffness at rest. Those with the highest EDP/EDV ratio had hyperdynamic LV systolic function and the smallest LV size. More work is needed to understand contribution of coronary microvascular dysfunction to HFpEF progression.

Funding Acknowledgement

Type of funding sources: Other. Main funding source(s): National Institutes of HealthErika Glazer Women's Heart Health Project

Unsupervised machine learning improves risk stratification of patients with visual normal SPECT myocardial perfusion imaging assessments

Background

Unsupervised machine learning has the potential to identify new cardiovascular phenotypes and more accurately assess individual risk in an unbiased fashion.

Purpose

We aimed to use unsupervised learning to identify, analyze, and risk-stratify subgroups of patients with normal perfusion by visual interpretation on single-photon emission computed tomography (SPECT) myocardial perfusion imaging (MPI).

Methods

We included consecutive patients with visual normal clinical assessment (summed stress score of 0) from the multicenter (9 sites), REFINE SPECT registry. We considered 23 clinical, 17 image-acquisition, and 26 imaging variables. Optimal dimensionality reduction (Uniform Manifold Approximation and Projection), clustering (Gaussian Mixture Model), and number of clusters were selected to maximize the silhouette coefficient (how similar a patient is to those in their own cluster compared to other clusters). Risk stratification for all-cause mortality (ACM) and major adverse cardiac events (MACE) was assessed within these clusters and compared to risk stratification by quantitative ischemia (&lt;5%, 5–10%, &gt;10%) using Kaplan-Meier curves and Cox Proportional-Hazards analysis.

Results

In total, 17,527 (of 30,351) patients in the registry had visually normal perfusion, 49.7% female, median age of 64 [55, 72] years. There were 1,138 ACM events and 2,091 MACE events with a median follow-up of 4.1 [2.9, 5.7] years. Unsupervised learning provided better risk stratification for both ACM and MACE compared to quantitative ischemia (Figure). Notably, the high-risk cluster by unsupervised learning had a hazard ratio (HR) of 9.5 (95% confidence interval [CI]: 7.7–11.7) compared to 1.4 (95% CI: 1.1–1.9) for quantitative ischemia &gt;10%. The high-risk cluster had proportionally more women (45% [low-risk], 51% [medium-risk], 57% [high-risk], all p&lt;0.001), higher body mass indices (26.9, 27.4, 29.6, all p&lt;0.001), prevalence of diabetes (17%, 22%, 33%, all p&lt;0.001), and abnormal rest ECGs (30%, 43%, 64%, p&lt;0.001); with lower rates of family history of coronary artery disease (40%, 33%, 24%, p&lt;0.001). Patients in the low-risk cluster were more likely to undergo exercise stress (100%, 38%, 0%, all p&lt;0.001), had lower rest peak systolic blood pressure (130, 131, 140 mmHg, all p&lt;0.001), and higher stress peak systolic blood pressure (164, 150, 131 mmHg, all p&lt;0.001). Patients in the high-risk cluster had higher left ventricular mass (129, 135.45, 143.9 g, all p&lt;0.001) and stress volume (57, 59, 66 ml, all p&lt;0.001).

Conclusion

Unsupervised learning identified new phenotypic clusters for SPECT MPI patients with visual normal assessments which provided improved risk stratification for ACM and MACE compared to SPECT ischemia. Such individualized risk assessment may allow better targeted management of patients with visually normal perfusion.

Funding Acknowledgement

Type of funding sources: Public Institution(s). Main funding source(s): Research reported in this publication was supported by the National Heart, Lung, And Blood Institute of the National Institutes of Health under Award Number R01HL089765. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.

Pericardial fat from a single horizontal long axis cardiac magnetic resonance cine image: a validation study against three-dimensional cardiac computed tomography

Background

Excess fat accumulation around the heart (i.e., pericardial fat) is positively associated with cardiovascular disease. Pericardial fat is composed of two distinct depots: (1) epicardial fat, which is a metabolically active adipose tissue located between the heart and the pericardium, and (2) paracardial fat, which is the fat deposit in the mediastinum outside of the parietal pericardium. Both depots are visible on the horizontal long axis cardiac magnetic resonance (CMR) cine (i.e., four chamber view), offering an attractive opportunity to quantify pericardial fat from standard CMR cine images.

Purpose

To validate pericardial fat area measured from a single horizontal long axis CMR cine against whole-heart volumetric non-contrast cardiac computed tomography (CT) measurements.

Methods

To accomplish our goal, we leveraged 25 cases from the Women's Ischemia Syndrome Evaluation – Coronary Vascular Dysfunction Continuation cohort who underwent both cardiac MRI and cardiac CT within a median of 35 days apart. For MRI, pericardial fat area was measured from a single high resolution steady state free precession cine image in the horizontal long axis imaging plane using commercially available software (CVI42 V5.13.5, Circle Cardiovascular Imaging, Figure 1A). For CT, pericardial fat volume was measured using a fully automated deep learning algorithm (QFAT 2.0, Figure 1A).

Results

Fat area measured from a single horizontal long axis cine image was closely related to fat volume measured by three-dimensional cardiac CT, with strong correlations for epicardial fat (R2=0.72, p&lt;0.01, Figure 1B), paracardial fat (R2=0.80, p&lt;0.01, Figure 1C), and pericardial fat (R2=0.91, p&lt;0.01, Figure 1D).

Conclusions

Measuring pericardial fat area, and its constituent parts, from a single horizontal long axis cine image is both feasible and strongly related to reference standard pericardial fat volume by cardiac CT.

Funding Acknowledgement

Type of funding sources: Public Institution(s). Main funding source(s): National Institutes of Health

Machine learning to predict abnormal myocardial perfusion from pre-test features

BACKGROUND

Accurately predicting which patients will have abnormal perfusion on MPI based on pre-test clinical information may help physicians make test selection decisions. We developed and validated a machine learning (ML) model for predicting abnormal perfusion using pre-test features.

METHODS

We included consecutive patients who underwent SPECT MPI, with 20,418 patients from a multi-center (5 sites) international registry in the training population and 9019 patients (from 2 separate sites) in the external testing population. The ML (extreme gradient boosting) model utilized 30 pre-test features to predict the presence of abnormal myocardial perfusion by expert visual interpretation.

RESULTS

In external testing, the ML model had higher prediction performance for abnormal perfusion (area under receiver-operating characteristic curve [AUC] 0.762, 95% CI 0.750-0.774) compared to the clinical CAD consortium (AUC 0.689) basic CAD consortium (AUC 0.657), and updated Diamond-Forrester models (AUC 0.658, p < 0.001 for all). Calibration (validation of the continuous risk prediction) was superior for the ML model (Brier score 0.149) compared to the other models (Brier score 0.165 to 0.198, all p < 0.001).

CONCLUSION

ML can predict abnormal myocardial perfusion using readily available pre-test information. This model could be used to help guide physician decisions regarding non-invasive test selection.

Diagnostic safety of a machine learning-based automatic patient selection algorithm for stress-only myocardial perfusion SPECT

BACKGROUND

Stress-only myocardial perfusion imaging (MPI) markedly reduces radiation dose, scanning time, and cost. We developed an automated clinical algorithm to safely cancel unnecessary rest imaging with high sensitivity for obstructive coronary artery disease (CAD).

METHODS AND RESULTS

Patients without known CAD undergoing both MPI and invasive coronary angiography from REFINE SPECT were studied. A machine learning score (MLS) for prediction of obstructive CAD was generated using stress-only MPI and pre-test clinical variables. An MLS threshold with a pre-defined sensitivity of 95% was applied to the automated patient selection algorithm. Obstructive CAD was present in 1309/2079 (63%) patients. MLS had higher area under the receiver operator characteristic curve (AUC) for prediction of CAD than reader diagnosis and TPD (0.84 vs 0.70 vs 0.78, P < .01). An MLS threshold of 0.29 had superior sensitivity than reader diagnosis and TPD for obstructive CAD (95% vs 87% vs 87%, P < .01) and high-risk CAD, defined as stenosis of the left main, proximal left anterior descending, or triple-vessel CAD (sensitivity 96% vs 89% vs 90%, P < .01).

CONCLUSIONS

The MLS is highly sensitive for prediction of both obstructive and high-risk CAD from stress-only MPI and can be applied to a stress-first protocol for automatic cancellation of unnecessary rest imaging.

Machine Learning From Quantitative Coronary Computed Tomography Angiography Predicts Fractional Flow Reserve-Defined Ischemia and Impaired Myocardial Blood Flow

BACKGROUND

A pathophysiological interplay exists between plaque morphology and coronary physiology. Machine learning (ML) is increasingly being applied to coronary computed tomography angiography (CCTA) for cardiovascular risk stratification. We sought to assess the performance of a ML score integrating CCTA-based quantitative plaque features for predicting vessel-specific ischemia by invasive fractional flow reserve (FFR) and impaired myocardial blood flow (MBF) by positron emission tomography (PET).

METHODS

This post-hoc analysis of the PACIFIC trial (Prospective Comparison of Cardiac Positron Emission Tomography/Computed Tomography [CT]' Single Photon Emission Computed Tomography/CT Perfusion Imaging and CT Coronary Angiography with Invasive Coronary Angiography) included 208 patients with suspected coronary artery disease who prospectively underwent CCTA' [15O]H2O PET, and invasive FFR. Plaque quantification from CCTA was performed using semiautomated software. An ML algorithm trained on the prospective NXT trial (484 vessels) was used to develop a ML score for the prediction of ischemia (FFR≤0.80), which was then evaluated in 581 vessels from the PACIFIC trial. Thereafter, the ML score was applied for predicting impaired hyperemic MBF (≤2.30 mL/min per g) from corresponding PET scans. The performance of the ML score was compared with CCTA reads and noninvasive FFR derived from CCTA (FFRCT).

RESULTS

One hundred thirty-nine (23.9%) vessels had FFR-defined ischemia, and 195 (33.6%) vessels had impaired hyperemic MBF. For the prediction of FFR-defined ischemia, the ML score yielded an area under the receiver-operating characteristic curve of 0.92, which was significantly higher than that of visual stenosis grade (0.84; P<0.001) and comparable with that of FFRCT (0.93; P=0.34). Quantitative percent diameter stenosis and low-density noncalcified plaque volume had the greatest ML feature importance for predicting FFR-defined ischemia. When applied for impaired MBF prediction, the ML score exhibited an area under the receiver-operating characteristic curve of 0.80; significantly higher than visual stenosis grade (area under the receiver-operating characteristic curve 0.74; P=0.02) and comparable with FFRCT (area under the receiver-operating characteristic curve 0.77; P=0.16).

CONCLUSIONS

An externally validated ML score integrating CCTA-based quantitative plaque features accurately predicts FFR-defined ischemia and impaired MBF by PET, performing superiorly to standard CCTA stenosis evaluation and comparably to FFRCT.

Deep Learning of Coronary Calcium Scores From PET/CT Attenuation Maps Accurately Predicts Adverse Cardiovascular Events

BACKGROUND

Assessment of coronary artery calcium (CAC) by computed tomographic (CT) imaging provides an accurate measure of atherosclerotic burden. CAC is also visible in computed tomographic attenuation correction (CTAC) scans, always acquired with cardiac positron emission tomographic (PET) imaging.

OBJECTIVES

The aim of this study was to develop a deep-learning (DL) model capable of fully automated CAC definition from PET CTAC scans.

METHODS

The novel DL model, originally developed for video applications, was adapted to rapidly quantify CAC. The model was trained using 9,543 expert-annotated CT scans and was tested in 4,331 patients from an external cohort undergoing PET/CT imaging with major adverse cardiac events (MACEs) (follow-up 4.3 years), including same-day paired electrocardiographically gated CAC scans available in 2,737 patients. MACE risk stratification in 4 CAC score categories (0, 1-100, 101-400, and >400) was analyzed and CAC scores derived from electrocardiographically gated CT scans (standard scores) by expert observers were compared with automatic DL scores from CTAC scans.

RESULTS

Automatic DL scoring required <6 seconds per scan. DL CTAC scores provided stepwise increase in the risk for MACE across the CAC score categories (HR up to 3.2; P < 0.001). Net reclassification improvement of standard CAC scores over DL CTAC scores was nonsignificant (-0.02; 95% CI: -0.11 to 0.07). The negative predictive values for MACE of zero CAC with standard (85%) and DL CTAC (83%) CAC scores were similar (P = 0.19).

CONCLUSIONS

DL CTAC scores predict cardiovascular risk similarly to standard CAC scores quantified manually by experienced operators from dedicated electrocardiographically gated CAC scans and can be obtained almost instantly, with no changes to PET/CT scanning protocol.

Deep Learning for Explainable Estimation of Mortality Risk From Myocardial Positron Emission Tomography Images

BACKGROUND

We aim to develop an explainable deep learning (DL) network for the prediction of all-cause mortality directly from positron emission tomography myocardial perfusion imaging flow and perfusion polar map data and evaluate it using prospective testing.

METHODS

A total of 4735 consecutive patients referred for stress and rest ⁸²Rb positron emission tomography between 2010 and 2018 were followed up for all-cause mortality for 4.15 (2.24-6.3) years. DL network utilized polar maps of stress and rest perfusion, myocardial blood flow, myocardial flow reserve, and spill-over fraction combined with cardiac volumes, singular indices, and sex. Patients scanned from 2010 to 2016 were used for training and validation. The network was tested in a set of 1135 patients scanned from 2017 to 2018 to simulate prospective clinical implementation.

RESULTS

In prospective testing, the area under the receiver operating characteristic curve for all-cause mortality prediction by DL (0.82 [95% CI, 0.77-0.86]) was higher than ischemia (0.60 [95% CI, 0.54-0.66]; P <0.001), myocardial flow reserve (0.70 [95% CI, 0.64-0.76], P <0.001) or a comprehensive logistic regression model (0.75 [95% CI, 0.69-0.80], P <0.05). The highest quartile of patients by DL had an annual all-cause mortality rate of 11.87% and had a 16.8 ([95% CI, 6.12%-46.3%]; P <0.001)-fold increase in the risk of death compared with the lowest quartile patients. DL showed a 21.6% overall reclassification improvement as compared with established measures of ischemia.

CONCLUSIONS

The DL model trained directly on polar maps allows improved patient risk stratification in comparison with established methods for positron emission tomography flow or perfusion assessments.

Early versus late acute coronary syndrome risk patterns of coronary atherosclerotic plaque

AIMS

The temporal instability of coronary atherosclerotic plaque preceding an incident acute coronary syndrome (ACS) is not well defined. We sought to examine differences in the volume and composition of coronary atherosclerosis between patients experiencing an early (≤90 days) versus late ACS (>90 days) after baseline coronary computed tomography angiography (CCTA).

METHODS AND RESULTS

From a multicenter study, we enrolled patients who underwent a clinically indicated baseline CCTA and experienced ACS during follow-up. Separate core laboratories performed blinded adjudication of ACS events and quantification of CCTA including compositional plaque volumes by Hounsfield units (HU): calcified plaque >350 HU, fibrous plaque 131-350 HU, fibrofatty plaque 31-130 HU and necrotic core <30 HU. In 234 patients (mean age 62 ± 12 years, 36% women), early and late ACS occurred in 129 and 105 patients after a mean of 395 ± 622 days, respectively. Patients with early ACS had a greater maximal diameter stenosis and maximal cross-sectional plaque burden as compared to patients with late ACS (P < 0.05). Larger total, fibrous, fibrofatty, and necrotic core volumes were observed in the early ACS group (P < 0.05). Findings for total, fibrous, fibrofatty, and necrotic core volumes were reproduced in an external validation cohort (P < 0.05).

CONCLUSIONS

Volumetric differences in composition of coronary atherosclerosis exist between ACS patients according to their timing antecedent to the acute event. These data support that a large burden of non-calcified plaque on CCTA is strongly associated with near-term plaque instability and ACS risk.

Determining a minimum set of variables for machine learning cardiovascular event prediction: results from REFINE SPECT registry

AIMS

Optimal risk stratification with machine learning (ML) from myocardial perfusion imaging (MPI) includes both clinical and imaging data. While most imaging variables can be derived automatically, clinical variables require manual collection, which is time-consuming and prone to error. We determined the fewest manually input and imaging variables required to maintain the prognostic accuracy for major adverse cardiac events (MACE) in patients undergoing a single-photon emission computed tomography (SPECT) MPI.

METHODS AND RESULTS

This study included 20 414 patients from the multicentre REFINE SPECT registry and 2984 from the University of Calgary for training and external testing of the ML models, respectively. ML models were trained using all variables (ML-All) and all image-derived variables (including age and sex, ML-Image). Next, ML models were sequentially trained by incrementally adding manually input and imaging variables to baseline ML models based on their importance ranking. The fewest variables were determined as the ML models (ML-Reduced, ML-Minimum, and ML-Image-Reduced) that achieved comparable prognostic performance to ML-All and ML-Image. Prognostic accuracy of the ML models was compared with visual diagnosis, stress total perfusion deficit (TPD), and traditional multivariable models using area under the receiver-operating characteristic curve (AUC). ML-Minimum (AUC 0.798) obtained comparable prognostic accuracy to ML-All (AUC 0.799, P = 0.19) by including 12 of 40 manually input variables and 11 of 58 imaging variables. ML-Reduced achieved comparable accuracy (AUC 0.796) with a reduced set of manually input variables and all imaging variables. In external validation, the ML models also obtained comparable or higher prognostic accuracy than traditional multivariable models.

CONCLUSION

Reduced ML models, including a minimum set of manually collected or imaging variables, achieved slightly lower accuracy compared to a full ML model but outperformed standard interpretation methods and risk models. ML models with fewer collected variables may be more practical for clinical implementation.

Subclinical hepatic fibrosis is associated with coronary microvascular dysfunction by myocardial perfusion reserve index: a retrospective cohort study

The heart-liver axis is of growing importance. Previous studies have identified independent association of liver dysfunction and fibrosis with adverse cardiac outcomes, but mechanistic pathways remain uncertain. We sought to understand the relations between the degree of hepatic fibrosis identified by the Fibrosis-4 (Fib-4) risk score and comprehensive cardiac MRI (CMR) measures of subclinical cardiac disease. We conducted a retrospective single-center cohort study of patients between 2011 and 2021. We identified consecutive patients who underwent a comprehensive CMR imaging protocol including contrast enhanced with stress/rest perfusion, and lacked pre-existing cardiovascular disease or perfusion abnormalities on CMR. We examined the association of hepatic fibrosis, using the Fib-4 score, with subclinical cardiac disease on CMR while adjusting for cardiometabolic traits. Given known associations of hepatic disease and coronary microvascular dysfunction, we prioritized analyses with the myocardial perfusion reserve index (MPRI), a marker of coronary microvascular function. Of the 66 patients in our study cohort, 54 were female (81%) and the mean age was 53.7 ± 15.3 years. We found that higher Fib-4 was associated with reduction in the MPRI (β [SE] - 1.12 [0.46], P = 0.02), after adjusting for cardiometabolic risk factors. Importantly, Fib-4 was not significantly associated with any other CMR phenotypes including measures of cardiac remodeling, inflammation, fibrosis, or dysfunction. We found evidence that hepatic fibrosis associated with coronary microvascular dysfunction, in the absence of overt associations with any other subclinical cardiac disease measures. These findings highlight a potentially important precursor pathway leading to development of subsequent heart-liver disease.

Coronary Artery Calcium for Risk Stratification of Sudden Cardiac Death: The Coronary Artery Calcium Consortium

BACKGROUND

Coronary artery calcium (CAC) is a marker of plaque burden. Whether CAC improves risk stratification for incident sudden cardiac death (SCD) beyond atherosclerotic cardiovascular disease (ASCVD) risk factors is unknown.

OBJECTIVES

SCD is a common initial manifestation of coronary heart disease (CHD); however, SCD risk prediction remains elusive.

METHODS

The authors studied 66,636 primary prevention patients from the CAC Consortium. Multivariable competing risks regression and C-statistics were used to assess the association between CAC and SCD, adjusting for demographics and traditional risk factors.

RESULTS

The mean age was 54.4 years, 33% were women, 11% were of non-White ethnicity, and 55% had CAC >0. A total of 211 SCD events (0.3%) were observed during a median follow-up of 10.6 years, 91% occurring among those with baseline CAC >0. Compared with CAC = 0, there was a stepwise higher risk (P trend < 0.001) in SCD for CAC 100 to 399 (subdistribution hazard ratio [SHR]: 2.8; 95% CI: 1.6-5.0), CAC 400 to 999 (SHR: 4.0; 95% CI: 2.2-7.3), and CAC >1,000 (SHR: 4.9; 95% CI: 2.6-9.9). CAC provided incremental improvements in the C-statistic for the prediction of SCD among individuals with a 10-year risk <7.5% (ΔC-statistic = +0.046; P = 0.02) and 7.5% to 20% (ΔC-statistic = +0.069; P = 0.003), which were larger when compared with persons with a 10-year risk >20% (ΔC-statistic = +0.01; P = 0.54).

CONCLUSIONS

Higher CAC burden strongly associates with incident SCD beyond traditional risk factors, particularly among primary prevention patients with low-intermediate risk. SCD risk stratification can be useful in the early stages of CHD through the measurement of CAC, identifying patients most likely to benefit from further downstream testing.

Longitudinal quantitative assessment of coronary atherosclerosis related to normal systolic blood pressure maintenance in the absence of established cardiovascular disease

BACKGROUND

Atherosclerosis-related adverse events are commonly observed even in conditions with low cardiovascular (CV) risk. Longitudinal data regarding the association of normal systolic blood pressure maintenance (SBP_maintain ) with coronary plaque volume changes (PVC) has been limited in adults without traditional CV disease.

HYPOTHESIS

Normal SBP_maintain is important to attenuate coronary atherosclerosis progression in adults without baseline CV disease.

METHODS

We analyzed 95 adults (56.7 ± 8.5 years; 40.0% men) without baseline CV disease who underwent serial coronary computed tomographic angiography with mean 3.5 years of follow-up. All participants were divided into two groups of normal SBP_maintain (follow-up SBP < 120 mm Hg) and ≥elevated SBP_maintain (follow-up SBP ≥ 120 mm Hg). Annualized PVC was defined as PVC divided by the interscan period.

RESULTS

Compared to participants with normal SBP_maintain , those with ≥elevated SBP_maintain had higher annualized total PVC (mm³ /year) (0.0 [0.0-2.2] vs. 4.1 [0.0-13.0]; p < .001). Baseline total plaque volume (β = .10) and the levels of SBP_maintain (β = .23) and follow-up high-density lipoprotein cholesterol (β = -0.28) were associated with annualized total PVC (all p < .05). The optimal cutoff of SBP_maintain for predicting plaque progression was 118.5 mm Hg (sensitivity: 78.2%, specificity: 62.5%; area under curve: 0.700; 95% confidence interval [CI]: 0.59-0.81; p < .05). SBP_maintain  ≥ 118.5 mm Hg (odds ratio [OR]: 4.03; 95% CI: 1.51-10.75) and baseline total plaque volume (OR: 1.03; 95% CI: 1.01-1.06) independently influenced coronary plaque progression (all p < .05).

CONCLUSION

Normal SBP_maintain is substantial to attenuate coronary atherosclerosis progression in conditions without established CV disease.

Differences in Prognostic Value of Myocardial Perfusion Single-Photon Emission Computed Tomography Using High-Efficiency Solid-State Detector Between Men and Women in a Large International Multicenter Study

BACKGROUND

Semiquantitative assessment of stress myocardial perfusion defect has been shown to have greater prognostic value for prediction of major adverse cardiac events (MACE) in women compared with men in single-center studies with conventional single-photon emission computed tomography (SPECT) cameras. We evaluated sex-specific difference in the prognostic value of automated quantification of ischemic total perfusion defect (ITPD) and the interaction between sex and ITPD using high-efficiency SPECT cameras with solid-state detectors in an international multicenter imaging registry (REFINE SPECT [Registry of Fast Myocardial Perfusion Imaging With Next-Generation SPECT]).

METHODS

Rest and exercise or pharmacological stress SPECT myocardial perfusion imaging were performed in 17 833 patients from 5 centers. MACE was defined as the first occurrence of death or myocardial infarction. Total perfusion defect (TPD) at rest, stress, and ejection fraction were quantified automatically by software. ITPD was given by stressTPD-restTPD. Cox proportional hazards model was used to evaluate the association between ITPD versus MACE-free survival and expressed as a hazard ratio.

RESULTS

In 10614 men and 7219 women, with a median follow-up of 4.75 years (interquartile range, 3.7-6.1), there were 1709 MACE. In a multivariable Cox model, after adjusting for revascularization and other confounding variables, ITPD was associated with MACE (hazard ratio, 1.08 [95% CI, 1.05-1.1]; P<0.001). There was an interaction between ITPD and sex (P<0.001); predicted survival for ITPD<5% was worse among men compared to women, whereas survival among women was worse than men for ITPD≥5%, P<0.001.

CONCLUSIONS

In the international, multicenter REFINE SPECT registry, moderate and severe ischemia as quantified by ITPD from high-efficiency SPECT is associated with a worse prognosis in women compared with men.

Longitudinal Quantitative Assessment of Coronary Atherosclerotic Plaque Burden Related to Serum Hemoglobin Levels

Background

Despite a potential role of hemoglobin in atherosclerosis, data on coronary plaque volume changes (PVC) related to serum hemoglobin levels are limited.

Objectives

The authors sought to evaluate coronary atherosclerotic plaque burden changes related to serum hemoglobin levels using serial coronary computed tomographic angiography (CCTA).

Methods

A total of 830 subjects (age 61 ± 10 years, 51.9% male) who underwent serial CCTA were analyzed. The median interscan period was 3.2 (IQR: 2.5-4.4) years. Quantitative assessment of coronary plaques was performed at both scans. All participants were stratified into 4 groups based on the quartile of baseline hemoglobin levels. Annualized total PVC (mm³/year) was defined as total PVC divided by the interscan period.

Results

Baseline total plaque volume (mm³) was not different among all groups (group I [lowest]: 34.1 [IQR: 0.0-127.4] vs group II: 28.8 [IQR: 0.0-123.0] vs group III: 49.9 [IQR: 5.6-135.0] vs group IV [highest]: 34.3 [IQR: 0.0-130.7]; P = 0.235). During follow-up, serum hemoglobin level changes (Δ hemoglobin; per 1 g/dL) was related to annualized total PVC (β = −0.114) in overall participants (P < 0.05). After adjusting for age, sex, traditional risk factors, baseline hemoglobin and creatinine levels, baseline total plaque volume, and the use of aspirin, beta-blocker, angiotensin-converting enzyme inhibitor or angiotensin receptor blocker, and statin, Δ hemoglobin significantly affected annualized total PVC in only the composite of groups I and II (β = −2.401; P = 0.004).

Conclusions

Serial CCTA findings suggest that Δ hemoglobin has an independent effect on coronary atherosclerosis. This effect might be influenced by baseline hemoglobin levels. (Progression of Atherosclerotic Plaque Determined by Computed Tomographic Angiography Imaging [PARADIGM]; NCT02803411)

Causes of cardiovascular and noncardiovascular death in the ISCHEMIA trial

BACKGROUND

The International Study of Comparative Health Effectiveness with Medical and Invasive Approaches trial demonstrated no overall difference in the composite primary endpoint and the secondary endpoints of cardiovascular (CV) death/myocardial infarction or all-cause mortality between an initial invasive or conservative strategy among participants with chronic coronary disease and moderate or severe myocardial ischemia. Detailed cause-specific death analyses have not been reported.

METHODS

We compared overall and cause-specific death rates by treatment group using Cox models with adjustment for pre-specified baseline covariates. Cause of death was adjudicated by an independent Clinical Events Committee as CV, non-CV, and undetermined. We evaluated the association of risk factors and treatment strategy with cause of death.

RESULTS

Four-year cumulative incidence rates for CV death were similar between invasive and conservative strategies (2.6% vs 3.0%; hazard ratio [HR] 0.98; 95% CI [0.70-1.38]), but non-CV death rates were higher in the invasive strategy (3.3% vs 2.1%; HR 1.45 [1.00-2.09]). Overall, 13% of deaths were attributed to undetermined causes (38/289). Fewer undetermined deaths (0.6% vs 1.3%; HR 0.48 [0.24-0.95]) and more malignancy deaths (2.0% vs 0.8%; HR 2.11 [1.23-3.60]) occurred in the invasive strategy than in the conservative strategy.

CONCLUSIONS

In International Study of Comparative Health Effectiveness with Medical and Invasive Approaches, all-cause and CV death rates were similar between treatment strategies. The observation of fewer undetermined deaths and more malignancy deaths in the invasive strategy remains unexplained. These findings should be interpreted with caution in the context of prior studies and the overall trial results.

Handling missing values in machine learning to predict patient-specific risk of adverse cardiac events: Insights from REFINE SPECT registry

BACKGROUND

Machine learning (ML) models can improve prediction of major adverse cardiovascular events (MACE), but in clinical practice some values may be missing. We evaluated the influence of missing values in ML models for patient-specific prediction of MACE risk.

METHODS

We included 20,179 patients from the multicenter REFINE SPECT registry with MACE follow-up data. We evaluated seven methods for handling missing values: 1) removal of variables with missing values (ML-Remove), 2) imputation with median and unique category for continuous and categorical variables, respectively (ML-Traditional), 3) unique category for missing variables (ML-Unique), 4) cluster-based imputation (ML-Cluster), 5) regression-based imputation (ML-Regression), 6) missRanger imputation (ML-MR), and 7) multiple imputation (ML-MICE). We trained ML models with full data and simulated missing values in testing patients. Prediction performance was evaluated using area under the receiver-operating characteristic curve (AUC) and compared with a model without missing values (ML-All), expert visual diagnosis and total perfusion deficit (TPD).

RESULTS

During mean follow-up of 4.7 ± 1.5 years, 3,541 patients experienced at least one MACE (3.7% annualized risk). ML-All (reference model-no missing values) had AUC 0.799 for MACE risk prediction. All seven models with missing values had lower AUC (ML-Remove: 0.778, ML-MICE: 0.774, ML-Cluster: 0.771, ML-Traditional: 0.771, ML-Regression: 0.770, ML-MR: 0.766, and ML-Unique: 0.766; p < 0.01 for ML-Remove vs remaining methods). Stress TPD (AUC 0.698) and visual diagnosis (0.681) had the lowest AUCs.

CONCLUSION

Missing values reduce the accuracy of ML models when predicting MACE risk. Removing variables with missing values and retraining the model may yield superior patient-level prediction performance.

Clinical Deployment of Explainable Artificial Intelligence of SPECT for Diagnosis of Coronary Artery Disease

BACKGROUND

Explainable artificial intelligence (AI) can be integrated within standard clinical software to facilitate the acceptance of the diagnostic findings during clinical interpretation.

OBJECTIVES

This study sought to develop and evaluate a novel, general purpose, explainable deep learning model (coronary artery disease-deep learning [CAD-DL]) for the detection of obstructive CAD following single-photon emission computed tomography (SPECT) myocardial perfusion imaging (MPI).

METHODS

A total of 3,578 patients with suspected CAD undergoing SPECT MPI and invasive coronary angiography within a 6-month interval from 9 centers were studied. CAD-DL computes the probability of obstructive CAD from stress myocardial perfusion, wall motion, and wall thickening maps, as well as left ventricular volumes, age, and sex. Myocardial regions contributing to the CAD-DL prediction are highlighted to explain the findings to the physician. A clinical prototype was integrated using a standard clinical workstation. Diagnostic performance by CAD-DL was compared to automated quantitative total perfusion deficit (TPD) and reader diagnosis.

RESULTS

In total, 2,247 patients (63%) had obstructive CAD. In 10-fold repeated testing, the area under the receiver-operating characteristic curve (AUC) (95% CI) was higher according to CAD-DL (AUC: 0.83 [95% CI: 0.82-0.85]) than stress TPD (AUC: 0.78 [95% CI: 0.77-0.80]) or reader diagnosis (AUC: 0.71 [95% CI: 0.69-0.72]; P < 0.0001 for both). In external testing, the AUC in 555 patients was higher according to CAD-DL (AUC: 0.80 [95% CI: 0.76-0.84]) than stress TPD (AUC: 0.73 [95% CI: 0.69-0.77]) or reader diagnosis (AUC: 0.65 [95% CI: 0.61-0.69]; P < 0.001 for all). The present model can be integrated within standard clinical software and generates results rapidly (<12 seconds on a standard clinical workstation) and therefore could readily be incorporated into a typical clinical workflow.

CONCLUSIONS

The deep-learning model significantly surpasses the diagnostic accuracy of standard quantitative analysis and clinical visual reading for MPI. Explainable artificial intelligence can be integrated within standard clinical software to facilitate acceptance of artificial intelligence diagnosis of CAD following MPI.

Relationship Between Coronary Artery Calcium and Atherosclerosis Progression Among Patients With Suspected Coronary Artery Disease

BACKGROUND

Among symptomatic patients, it remains unclear whether a coronary artery calcium (CAC) score alone is sufficient or misses a sizeable burden and progressive risk associated with obstructive and nonobstructive atherosclerotic plaque.

OBJECTIVES

Among patients with low to high CAC scores, our aims were to quantify co-occurring obstructive and nonobstructive noncalcified plaque and serial progression of atherosclerotic plaque volume.

METHODS

A total of 698 symptomatic patients with suspected coronary artery disease (CAD) underwent serial coronary computed tomographic angiography (CTA) performed 3.5 to 4.0 years apart. Atherosclerotic plaque was quantified, including by compositional subgroups. Obstructive CAD was defined as ≥50% stenosis. Multivariate linear regression models were used to measure atherosclerotic plaque progression by CAC scores. Cox proportional hazard models estimated CAD event risk (median of 10.7 years of follow-up).

RESULTS

Across baseline CAC scores from 0 to ≥400, total plaque volume ranged from 30.4 to 522.4 mm³ (P < 0.001) and the prevalence of obstructive CAD increased from 1.4% to 49.1% (P < 0.001). Of those with a 0 CAC score, 97.9% of total plaque was noncalcified. Among patients with baseline CAC <100, nonobstructive CAD was prevalent (40% and 89% in CAC scores of 0 and 1-99), with plaque largely being noncalcified. On the follow-up coronary CTA, volumetric plaque growth (P < 0.001) and the development of new or worsening stenosis (P < 0.001) occurred more among patients with baseline CAC ≥100. Progression varied compositionally by baseline CAC scores. Patients with no CAC had disproportionate growth in noncalcified plaque, and for every 1 mm³ increase in calcified plaque, there was a 5.5 mm³ increase in noncalcified plaque volume. By comparison, patients with CAC scores of ≥400 exhibited disproportionate growth in calcified plaque with a volumetric increase 15.7-fold that of noncalcified plaque. There was a graded increase in CAD event risk by the CAC with rates from 3.3% for no CAC to 21.9% for CAC ≥400 (P < 0.001).

CONCLUSIONS

CAC imperfectly characterizes atherosclerotic disease burden, but its subgroups exhibit pathogenic patterns of early to advanced disease progression and stratify long-term prognostic risk.