Skip main navigation

Radiogenomics and Artificial Intelligence Approaches Applied to Cardiac Computed Tomography Angiography and Cardiac Magnetic Resonance for Precision Medicine in Coronary Heart Disease: A Systematic Review

Originally publishedhttps://doi.org/10.1161/CIRCIMAGING.121.013025Circulation: Cardiovascular Imaging. 2021;14:1133–1146

Abstract

The risk of coronary heart disease (CHD) clinical manifestations and patient management is estimated according to risk scores accounting multifactorial risk factors, thus failing to cover the individual cardiovascular risk. Technological improvements in the field of medical imaging, in particular, in cardiac computed tomography angiography and cardiac magnetic resonance protocols, laid the development of radiogenomics. Radiogenomics aims to integrate a huge number of imaging features and molecular profiles to identify optimal radiomic/biomarker signatures. In addition, supervised and unsupervised artificial intelligence algorithms have the potential to combine different layers of data (imaging parameters and features, clinical variables and biomarkers) and elaborate complex and specific CHD risk models allowing more accurate diagnosis and reliable prognosis prediction. Literature from the past 5 years was systematically collected from PubMed and Scopus databases, and 60 studies were selected. We speculated the applicability of radiogenomics and artificial intelligence through the application of machine learning algorithms to identify CHD and characterize atherosclerotic lesions and myocardial abnormalities. Radiomic features extracted by cardiac computed tomography angiography and cardiac magnetic resonance showed good diagnostic accuracy for the identification of coronary plaques and myocardium structure; on the other hand, few studies exploited radiogenomics integration, thus suggesting further research efforts in this field. Cardiac computed tomography angiography resulted the most used noninvasive imaging modality for artificial intelligence applications. Several studies provided high performance for CHD diagnosis, classification, and prognostic assessment even though several efforts are still needed to validate and standardize algorithms for CHD patient routine according to good medical practice.

Despite the improvements in prevention and early disease intervention, coronary heart disease (CHD) remains leading cause of disability and death worldwide.1–3 The risk to develop clinical manifestations of CHD and the clinical decision-making is currently estimated using score algorithms where multifactorial cardiovascular risk factors are the predominant indicators. However, this approach still fails to capture the individual cardiovascular risk.4 Hence, the discovery of novel biomarkers and approaches aimed to improve risk prediction algorithms will be a key to accomplish the promise of personalized medicine.5

Different noninvasive imaging techniques are available in the clinical practice to detect and characterize CHD such as cardiac computed tomography angiography (CCTA), cardiac magnetic resonance (CMR).6–8 CCTA has gained a worldwide clinical acceptance for CHD evaluation and risk prediction for its ability to exclude coronary stenosis and visualize vessel wall abnormalities and plaque morphology/composition.9 Indeed, the detection of high-risk plaque markers allows for a highly specific stratification of patients with an increased risk for acute events even in the presence of nonobstructive atherosclerotic lesions.2,10–12

CMR provides measurements of markers of cardiac structure and function, myocardial perfusion and scar, as well as detailed insight of myocardial tissue.13 CMR findings were demonstrated to have prognostic values; in particular, left ventricular ejection fraction (LVEF) resulted an independent predictor of future cardiovascular events in patients with a recent myocardial infarction (MI), while wall motion abnormalities, inducible perfusion defects, and LVEF were predictors for patients with suspected or known CHD.14,15

Technological improvements allowed to depict radiological images in 3D datasets and to extract several quantitative features that could be used to accurately phenotype a given lesion. Radiomics is the process of extrapolating from a region of interest such great amount of parameters with the goal to investigate correlations between such quantitative variables and clinical data.16

Recently, the integration of imaging data with molecular markers provided promising insights for CHD detection and characterization.8,17–19 Together with radiomics development, innovations in next-generation sequencing field allowed also to generate a considerable amount of biological big data and big data analytics.20–22 Radiogenomics integrate a huge amount of features extracted from medical images with genomic phenotypes aimed to build prediction models through the elaboration of complex algorithms aimed to stratify patients, guide accurate therapeutic regimens, and evaluate clinical outcomes.23–25 In this scenario, artificial intelligence (AI) is increasingly applied in research for disease definition and risk prediction26,27 (Figure 1). AI describes the use of computational techniques such as machine learning (ML) focused on the automatization of medical tasks to support clinicians in workflow optimization, diagnosis, and prognosis assessment in particular into imaging field.28 ML and deep learning (DL) are branches of AI, based on the development of algorithms able to learn without explicit instructions or programming by using a large amount of complex data.29,30 Therefore, ML has been widely applied in radiology and recently in the field of biology and bioinformatics.30,31 ML and DL combine different types of data such as medical imaging data and features, laboratory measurements, clinical variables and outcomes, biomarkers in neural networks aimed to perform an accurate patient stratification and prognostication toward a precision medicine approach.32 To improve radiogenomics and AI studies, integrated databases are available, sharing multiple data types ranging from clinical information, biological big data and cardiac radiological images (Table S1).

Figure 1.

Figure 1. Radiogenomics and artificial intelligence support to cardiovascular imaging analysis. Data acquisition generally includes computed tomography (CT) assessment of coronary anatomy and myocardial function, strengthened by the possibility of multiplanar reconstructions and angiographic views or the possibility to obtain DICOM and molecular data from large databases. Paralleling, the possibility to collect biological samples from the same patient, provides a unique framework for integrated radiogenomics studies. During imaging, pre- and post-processing on dedicated workstations is possible to finely segment plaque components (green for fibrotic, blue for lipidic, and yellow for the calcified, respectively); similarly, samples can be processed to extract plasma and serum parts or other nucleic/proteomic components. Data extraction represents the quantitative part of the pipeline allowing to achieve first, second, and third order features from radiomics algorithms and multiomics biological profiling/signatures by different samples. All the parameters can be analyzed mainly through machine learning (supervised and unsupervised) approaches, and the derived models validated through internal and external independent datasets to follow both diagnostic (patient classification) and prognostic purposes in coronary heart disease (CHD).

Nevertheless, these promising perspectives and strengths, there are also many drawbacks (Figure 2) and no clear guidelines to perform AI approaches in clinical studies.33 Thus, the aim of this systematic review is to argue the applicability of CCTA- and CMR-based radiomics, radiogenomics, and AI models for CHD diagnosis, characterization of atherosclerotic lesions and myocardial abnormalities, thus assessing the accuracy of such applications in the clinical practice.

Figure 2.

Figure 2. Strengths and limitations of artificial intelligence (AI) applied to cardiovascular imaging and biological datasets. CHD indicates coronary heart disease.

Methodology and Results

This systematic review was performed according to Preferred Reporting Items for Systematic Reviews and Meta-Analyses guidelines with a protocol agreed by all authors, without prospectively registering34 (Table S2). Two researchers, blinded for article author information, conducted the search, study inclusion, and data extraction. The search was performed on PubMed and Scopus databases with the following selection criteria: articles published in the last 5 years (from 2016 to 2021) study on adult humans and in English language. The specific search strategy and keywords for each search database are presented in Table S3. An author recovered the initial search results and removed duplicates and subsequently another researcher searched for the possible removal of other duplicates. Finally, 2 authors independently screened the studies by title, abstract, and keywords, then the selected studies were evaluated on the basis of inclusion and exclusion criteria (Table S4). From the literature search, 60 studies were selected. Radiomic features extracted by both CCTA and CMR studies showed good diagnostic accuracy for the identification of coronary plaques and myocardium structure; on the other hand, few studies exploited radiogenomics mainly in CCTA (n=4), thus suggesting the need for research efforts in this field. CCTA resulted in the most used noninvasive imaging modality for AI and ML applications and several studies (n=36) provided high performance values for CHD diagnosis, classification, disease phenotyping, and prognostic assessment as compared with AI-based CMR studies (n=10). Below, we will discuss in detail all the selected studies from our search.

Radiomics in CHD

Radiomics has the peculiarity to provide image characteristics not visible to the human eye. Radiomics has been largely investigated in oncology23,35,36 while research in cardiovascular imaging is still limited. Nevertheless, different studies demonstrated the feasibility and potential clinical value of radiomics analysis in CCTA and CMR.

Radiomics and AI are strictly interconnected due to the capability of AI to manage and analyze the massive amount of data extracted from medical images as compared with the traditional statistical methods. Although AI techniques have been primarily applied to build classification and predictive models on unseen datasets, AI computational algorithms are integrated in the radiomics workflow not only for the analyses of the extracted features but can directly analyze the images or perform segmentation tasks itself through DL algorithms.37 With the increasing use of noninvasive imaging technique for CHD assessment, the focus of AI research and applications has strongly expanded.

Radiomics in CCTA

Radiomics based CCTA studies aimed to provide volumetric and radiomic analysis for improving the diagnostic performance and better characterizing atherosclerotic plaques. In particular, literature data reported the robustness of diagnostic performance of radiomic models as compared with conventional CCTA features.38 In addition, CCTA-derived radiomic features were also effective to identify vulnerable plaques39 and to discriminate between Napkin-Ring Sign and non-Napkin-Ring Sign plaques with an area under the curve (AUC)>0.80.40

A radiomic analysis on 1103 parameters extracted from pericoronary adipose tissue scans of patients with stable CHD, acute MI, and controls was able to significantly differentiate patients with MI versus controls and patients with MI from stable CHD. The most significant radiomic parameters were texture or geometry based. At 6 months post-MI, there were no significant changes in pericoronary adipose tissue attenuation (PCAT) radiomic profile, thus suggesting a specific PCAT radiomic signature.41

In a retrospective study, Mannil and coauthors analyzed radiomic features extracted by noncontrast-enhanced ECG-gated CCTA scans to detect patients with MI from healthy subjects. The model generated provided an AUC of 0.78 to discriminate controls and patients.42

Radiomics in CMR

CMR findings are considered markers of prognosis following percutaneous coronary intervention after MI for the assessment of tissue revascularization, left ventricle (LV) remodeling, presence of scar/necrotic areas, and detection of small subendocardial MI.43

Baessler et al44 reported showed the accuracy of 5 independent texture features for the discrimination of subacute and chronic MI by using noncontrast-enhanced cine CMR. In addition, the combination of 2 features reported the highest accuracy for detecting large and small MI, with an AUC of 0.93 and 0.92, respectively.

Larroza et al45 applied 5 radiomic texture analysis to differentiate infarcted nonviable, viable, and remote cardiac segments in noncontrast-enhanced cine CMR reporting an AUC of 0.849 for the detection of nonviable segments, 0.85 for remote segments, and 0.72 for viable segments. The same group previously reported a radiomic analysis on late gadolinium enhancement (LGE) and standard cine CMR sequences in a cohort of patient with acute and chronic MI.46 A total of 279 features were extracted from predefined regions of interest whose classification performance was evaluated. On LGE CMR 72 features provided an AUC of 0.86 for acute MI detection, while on cine CMR analysis on 75 features retrieved an AUC of 0.82 for acute MI discrimination.46

Several radiomics-based CMR studies highlighted the prognostic value of texture analysis on quantitative extracellular volume fraction and T1 mapping to differentiate between reversible and irreversible myocardial damage and to predict adverse LV remodeling in patients with reperfused ST-segment elevation MI. In particular, 5 texture features were statistically significant to discriminate the extent of myocardial injury with an AUC of 0.91 for the differentiation of unsalvageable salvageable myocardium as well as LV adverse remodeling.47

Ma et al evaluated the added value of texture analysis on noncontrast-enhanced T1 mapping for the evaluation of myocardial injury after ST-segment elevation MI and the prediction cardiac functional recovery. The combination of radiomic features of T1 mapping and T1 values resulted in an incremental diagnostic value for the discrimination of positive microvascular obstruction with an AUC=0.86.

In addition, combined radiomics and T1 values were able to detect the presence of MI at different extent of transmurality and to predict irreversible segmental scar and LV segmental longitudinal strain at 6 months.48

A large study by using CMR dataset from 5065 individuals with known cardiovascular diseases and risk factors and healthy subjects from the UK Biobank aimed to perform a radiomic analysis of short-axis images of the LV and RV at end-diastole and end-systole. Texture analysis provided 684 features linked to shape, first-order and texture; 114 features were related to the conventional CMR indices such as LV end-diastolic volume, LV end-systolic volume, RV end-diastolic volume, RV end-systolic volume, LV stroke volume, RV stroke volume, LVEF, RV ejection fraction, LV mass (LVM). Specific radiomics signatures best characterized the structural and tissue differences between risk factor groups and healthy groups49 (Table 1).

Table 1. Radiomic Studies in CHD

Imaging protocolParticipantsExtracted feature categoriesResultsPerformanceReference
CCTA
 Retro- or prospective ECG-gated acquisition mode881359 (18 first-order statistics, 14 shape-based statistics, 75 texture features, 744 wavelet features, 508 filtered features)Diagnostic model built with 3 wavelet features showed a better performance for the discrimination of ischemic from nonischemic lesions as compared with conventional modelAUC=0.762 in training setHu et al38
AUC=0.671 in validation set
 Retrospective study27935 (44 first-order statistics, 342 GLCM statistics, 33 GLRLM, 516 geometry-based statistical parameters)CCTA radiomics showed a good diagnostic accuracy for the identification of vulnerable plaques compared with conventional methodsAUC range=0.72–0.87Kolossváry et al39
 Retrospective study26744440 (44 first-order statistics, 3585 GLCM based parameters, 55 GLRLM based metrics, 756 geometry-based statistics)Radiomic parameters allowed to better identifying NRS coronary lesions compared with conventional image reconstruction metrics. Short- and long-run low gray-level emphasis and surface ratio of high attenuation voxels to total surface showed the best diagnostic performanceAUC=0.918Kolossváry et al40
AUC=0.894
AUC=0.890
 Retrospective study601103 (44 first-order parameters, 342 GLCM parameters, 33 GLRM parameters, 684 geometric parameters describing shape, size/or volume)MI patients had a distinct PCAT radiomic phenotype as compared with healthy subjects and stable patients with CHD. Textural and geometric features were the most powerful in discriminating patients with MIAUC=0.87Lin et al41
 Noncontrast low-dose CCTA87308 (histogram, GLCM, run-length matrix at 4 angles, absolute gradient, autoregressive model, wavelet transform)TA features GLCM S5,-5InvDfMom, Teta2, and Teta3 enabled the differentiation between patients with MI and controlsAUC=0.78Mannil et al42
CMR
 Nonenhanced cine-CMR180206 (histogram, co-occurrence matrix, run-length matrix, absolute gradient, autoregressive mode, wavelet transform)Five independent TA features (Teta1, Perc.01, Variance, WavEnHH.s-3 and S(5,5) SumEntrp) allowed to discriminate subacute and chronic MI. Teta1 and Perc.01 resulted in the highest accuracy to diagnose large and small MIAUC=0.93Baessler et al44
AUC=0.92
 Cine and LGE CMR50Matrix, LBP features, matrix+LBP features, wall thickeningTA showed the potential to discriminate between nonviable, viable, and remote segments in MI patientsAUC=0.849Larroza et al45
 Cine and LGE CMR44279 (histogram, absolute gradient, GLCM, GLRLM, autoregressive model, wavelets)72 features extracted from LGE CMR and 75 features from cine CMR allowed to differentiate acute MI from chronic MIAUC=0.86 for LGE CMRLarroza et al46
AUC=0.82 for cine CMR
 Native and contrast T1 mapping CMR70279 (histogram indexes, co-occurrence matrix features S (0.1) difference entropy, and run-length matrix features)Five TA features extracted from ECV were able to identify the extent of myocardial injury and differentiate unsalvageable from salvageable myocardium as well as LV adverse remodeling in STEMIAUC=0.91Chen et al47
 T2-w, T1 mapping, rest first-pass perfusion, and LGE CMR68279 (histogram, absolute gradient, GLCM, GLRLM, autoregressive model, wavelets)Incremental diagnostic value of radiomic features for T1 mapping in distinguishing positive MVO. High accuracy of radiomic T1 values in the detection of MI at different extents of transmurality. Combined radiomics and T1 values predicted the irreversible segmental scar and recovery of LV SLS in STEMIAUC=0.86Ma et al48
AUC=0.77
 White blood CMR, cine CMR, strain CMR, flow CMR, native T1 mapping5065684 (shape, first-order, texture)Specific radiomics signatures best characterized myocardial structural and tissue differences between individuals with specific cardiovascular risk factors and healthy subjectsAUC range=0.63–0.80Cetin et al49

AUC indicates area under curve; CCTA, cardiac computed tomography angiography; CHD, coronary heart disease; CMR, cardiac magnetic resonance; ECV, extracellular volume; GLCM, gray-level co-occurrence matrix; GLRLM, gray-level run length matrix; LBP, local binary pattern; LGE, late gadolinium enhancement; LV, left ventricle; MI, myocardial infarction; MVO, microvascular obstruction; NRS, napkin-ring sign; PCAT, pericoronary adipose tissue; SLS, segmental longitudinal strain; STEMI, ST-segment elevation MI; and TA, texture analysis.

AI and Machine Learning Models in CHD

AI application in cardiac imaging aims to reduce the time of image analysis, thus facilitating the management of patients with CHD. Additionally, the integration of AI data in ML models comprising clinical and molecular data could provide an accurate diagnosis and prognostic stratification.29

ML algorithms can be technically divided into supervised and unsupervised. Supervised learning uses data that have been tagged with 1 or more labels, like properties, characteristics, or classifications, while unsupervised learning employs not tagged data.50 Furthermore, supervised algorithms is focused on classification of data into several subsets and prediction/estimation of unknown variables. Unsupervised algorithms are projected on the discovery of underlying patterns and relationships among the unlabelled data.50

AI-Based CCTA Applications

Several AI-based CCTA models have been proposed for CHD detection, evaluation of coronary stenosis, atherosclerotic plaque phenotyping, and quantification of myocardial ischemia.

CHD Detection

The rapid evaluation by CCTA of patients with chest pain in emergency department could be useful for a prompt intervention. In this regard, AI algorithms and workflow for supporting physicians in CCTA screening rule out CHD with AUC of 0.96 and a negative predictive value of 95%.51

A recent study by Han et al52 showed that CCTA-AI significantly reduced the time of image post-processing and accurately improved the identification of ≥50% coronary stenoses as compared with traditional CCTA analysis while there were a moderate accuracy for plaque classification. Three-dimensional convolutional neural network (CNN) models were found to be able to detect and characterize plaque composition and stenosis degrees.53,54 For identification and characterization of coronary plaques, 3D CNN method achieved an accuracy of 0.77 and for detection of stenosis and the method reached an accuracy of 0.80.53 In addition, AI convolutional autoencoders application also provided encouraging results for functionally significant stenosis identification requiring invasive coronary angiography with an accuracy of 0.87.55

van Hamersvelt et al56 developed an unsupervised algorithm for the detection of significant CHD using a combined approach integrating CCTA and invasive fractional flow reserve (FFR), which allowed to identify hemodynamically significant CHD with an AUC of 0.76. In addition, an unsupervised DL analysis of LV segmentation allowed the automatic classification of patients for the presence of functionally significant stenosis in one or more coronaries.57

Two studies used automated approaches based on CAD-RADS score.58,59 Muscogiuri et al58 generated a DL CNN for the classification of CAD-RADS developing 3 models with an accurate yield and short time of analysis as compared with human examination. Huang and coauthors classified CAD-RADS using a DL algorithm and correlated data with the mammary artery calcification assessed by screening mammography. Breast arterial calcification was significantly higher for high CAD-RADS scores, thus providing the useful opportunity to use the screening for breast cancer also for the early assessment of CHD.59

In a recent study, PCAT attenuation was integrated in a ML (XGBoost) model together with risk factors, serum lipids, high-sensitivity C-reactive protein, and specific radiomic features and provided superior discrimination of MI with an AUC of 0.87.41

A very recent investigation provided a high correlation of AI approach and the consensus of 3 readers for determination of % stenosis and CAD-RADS score, thus identifying a wide range of atherosclerotic plaque volume and plaque composition.60

CACS Evaluation and Coronary Plaque Classification

Raw data from coronary calcium score (CACS) scans were used to screen and predict the presence of obstructive CHD through the elaboration of a gradient boosting machine with a sensitivity of 100% and a specificity of 69.8% and a negative predictive value of 100%.61

Wang et al62 investigated the accuracy of a DL algorithm versus the classical quantification of CACS reporting and agreement between the 2 approaches with a correlation coefficient of 0.77, thus providing a reliable method for cardiac risk stratification.

Several studies performed comparisons between an automatic AI-based CACS ECG-gated CT post-processing software and traditional semiautomatic and manual softwares by evaluating the correlation and agreement of Agatson score (AS), volume score (VS), mass score (MS), and the number of calcified coronary lesions provided excellent correlation and agreement among the AI-automatic, the semiautomatic software and manual method for CACS indicators and the calcified lesion number with also less time consuming for analysis.63–65

Wolterink and colleagues proposed a supervised learning CNN method for automated CACS quantification without the need for coronary artery extraction. The algorithm was able to identify CAC with a sensitivity of 0.72 and an interclass correlation of 0.94 between CAC derived from CCTA and standard evaluation of CAC, thus avoiding the need to acquire a dedicated scan for CACS, which is regularly acquired before a CCTA and reducing radiation dose exposure.66

DL methods were also applied for the evaluation of the performance of automatic CACS assessment from different no contrast cardiac and chest CT scans. Despite substantial variation in CT protocols and subject population, the calcium scoring algorithm was robust for the quantification of coronary and thoracic aorta calcium.67,68

van den Oever et al69 developed an algorithm able to exclude CCTA negative scans and segments for CAC with an accuracy for correct classification of 86%, thus significantly reducing the workload of radiologists.

A recurrent neural network (RNN) with long short-term memory was used to automatically detect coronary calcified plaques from contrast-enhanced CCTA demonstrated that the algorithm had an high overall diagnostic performance for calcified lesion detection (sensitivity 92.1%; specificity 88.9%; accuracy 90.3%).70 In addition, Masuda et al71 applied a ML histogram for the identification of fibrous and fatty or mixed plaques compared with IVUS showing an accuracy of 0.92.

In addition to CAC measurement, noncontrast CCTA can be used for the quantification of epicardial adipose tissue (EAT). Routine measurement of EAT is time consuming. Hence, DL was used to automated quantified EAT volume and attenuation in 2068 subjects. EAT volume and attenuation predicted major adverse cardiovascular event (MACE) in asymptomatic subjects, independently by traditional risk factors and CACS. This novel metrics provided additional prognostic information extracted from noncontrast cardiac scans without physician interaction.72

ConvNet algorithm applied on a CT imaging data sets from 250 asymptomatic subjects, allowed to detect EAT with a strong agreement as compared with manual quantification with a dice similarity coefficient of 0.823 and correlation coefficient of 0.924.73 An unsupervised hierarchical clustering analysis was able to discriminate high-risk plaque features and increased pericoronary adipose tissue attenuation expressed by fat attenuation index.74

A recent study on asymptomatic 1069 subjects from the prospective EISNER trial undergoing CCTA determined the performance of a ML risk score CCTA derived parameters and circulating biomarkers for the prediction of long-term (14.5±2.0 years) risk of cardiac events. XGBoost algorithm was trained using traditional clinical variables, a wide panel of serum proteins and 5 quantitative CT parameters (CACS, number of calcified lesions, aortic valve calcium score, EAT volume, fat attenuation index). The ML score provided an AUC of 0.81, superior of CAC and ASCVD risk scores in the prediction of MI and cardiac death. In addition, serum biomarkers such as MMP-9 (matrix metallopeptidase 9), pentraxin 3, PIGR (polymeric immunoglobulin receptor), and GDF-15 (growth differentiation factor-15) significantly increased the prognostic value of the model.75

A ML-based targeted radiogenomic approach was used to predict high-risk plaque or absence of coronary atherosclerosis, assessed by CCTA, in suspected patients with CHD. Circulating levels of 358 proteins were considered to generate a training ML model; then, the performance was validated against a clinical model comprising clinical data and conventional biomarkers. The model generated had an AUC of 0.79, outperforming the conventional risk score. Specific protein signatures were identified. In particular, 35 plasma proteins were predictive of the presence of high-risk coronary plaques and a subset of 34 proteins was associated to absence of coronary lesions. Both the models generated had a superior performance as compared with conventional biomarkers.76

Oikonomou et al developed and validated a ML-derived radiotranscriptomic approach for cardiac risk assessment in 3 different study populations. On a group of 167 patients who underwent cardiac surgery, radiomic profile of PCAT was determined by CCTA.77 Transcriptome profile was determined on PCAT tissues analyzing genes linked to inflammation, fibrosis, and vascularity. Radiotranscriptomic analysis showed that wavelet-transformed mean attenuation (expressed by fat attenuation index) was the best-performing feature for the detection of adipose tissue inflammation, corroborated by the expression of TNFA in biopsy samples. In addition, texture features were related to fibrosis and vascularity (COL1A1 expression and CD31 expression). The radiogenomic signature was validated by analyzing 1391 coronary PCAT radiomic features in 101 patients with MACE within 5 years after CCTA and 101 control subjects and a random forest algorithm was built to discriminate cases with MACE and controls. The model was then tested in a group of 1575 consecutive patients undergoing CCTA providing a significant improved MACE prediction as compared with traditional risk classification. The fat radiomic phenotype was also investigated in 44 patients with MI scanned within 96 hours of hospitalization and 44 patients with stable CHD used as controls revealing higher fat radiomic phenotype values (corresponding to adverse PCAT remodeling) in the MI versus control group. After 6 months, no changes in fat radiomic phenotype values were observed while surrounding right coronary artery perivascular fat attenuation index, a specific biomarker of coronary inflammation, changed dynamically, thus providing information on the residual cardiac risk.77

Evaluation of Myocardial Ischemia

Recent research and development in AI has been applied for the evaluation of myocardial ischemia78 with a particular focus on FFR for the detection of hemodynamically significant CHD.79 An AI-based FFR CT, derived from triple rule out CT datasets, provided additional diagnostic and prognostic values in patients with acute chest pain, thus reducing the need of additional testing.80 A 3D DL model on fully automatic estimation of minimum FFR-CCTA data achieved an AUC of 0.78 for detection of abnormal FFR.81

The MACHINE consortium (Machine Learning Based CT Angiography Derived FFR: a Multi-Center Registry) applied a ML-based CT-FFR in 351 patients with 525 vessels reporting that the diagnostic accuracy of ML approach was superior to CCTA in identifying hemodynamically significant CHD (AUC=0.84)82 without sex difference.83 Other studies highlighted the significant predictive value of FFR ML algorithm for the assessment of lesion-specific ischemia and flow-limiting stenosis.84–86 ML-based CT-FFR was also useful to detect coronary calcification with a superior diagnostic value over conventional CCTA analysis in vessels with high and low-intermediate AS.87

DL-FFRCT reduced the need for diagnostic coronary angiography with a low MACE rate in a 2-year follow-up.88 In addition, ML-based FFRCT models were feasible for the evaluation of in-stent restenosis in patients with stent implantation showing prognostic values in predicting MACE after stent implantation89 (Table 2).

Table 2. CCTA-Based AI Applications in CHD

Imaging protocolParticipantsAI applicationResultsPerformanceReference
ECG-gated CCTA500Algorithm using MPVs and GUIAn AI algorithm was able to identify the absence of coronary atherosclerosis in subjects with chest painAUC=0.96White et al51
2D CT images reconstruction150DL and TLAI algorithm significantly reduced post-processing and CHD diagnosis and the detection of coronary stenosis ≥50%AUC=0.87Han et al52
AUC=0.75
CCTA163Recurrent 3D CNNIdentification of coronary plaque composition and stenosis degreeAUC=0.77 for plaqueZreik et al53
AUC=0.80 for stenosis
CCTA64CNNThe identification of hemodynamically significant CHD was not inferior to traditional methodsAUC=0.86Podgorsak et al54
CCTA187Unsupervised convolutional autoencodersAutomatic detection of functionally significant coronary stenosisAUC=0.87Zreik et al55
CCTA126CNNImproved diagnostic performance for the detection of functional stenosis by the combination of DL analysis of LVMAUC=0.76van Hamersvelt et al56
CCTA166Unsupervised convolutional autoencoder; CNNA DL algorithm applied on LV images allowed to automatic classify patients for functionally significant stenosisAUC = 0.74Zreik et al57
CCTA288Unsupervised CNNDevelopment of 3 CNN models for automated classification of patients with CHD according to CAD-RADSMuscogiuri et al58
CCTA213DLCorrelation between CHD severity, detected by DL, CAD-RADS, and mammary arterial calcification in women undergoing screening mammographyHuang et al59
ECG-gated CCTA232CNN models (including VGG 19 network, 3D U-Net and VGG Network Variant)AI showed high correlation to 3 reader consensus for % of stenosis and CAD-RADS scoreAccuracy, sensitivity, specificity (>70% stenosis: 99.7%, 90.9%, 99.8%; >50% stenosis: 94.8%, 80.0%, 97.0%)Choi et al60
CACS CT435GBMDevelopment of an AI model based on automatic CACS able to screen patients with obstructive CHDSensitivity=100%Głowacki et al61
Specificity=69.8%
CACS CT530DLDL algorithm provided reliable AS and volume scores as compared with traditional method, enabling cardiac risk stratificationCC=0.77Wang et al62
CACS CT315Prior likelihoodCorrelation between the automatic AI and the semiautomatic software for CACS and number of calcified lesionsCC=0.935 for ASSandstedt et al63
CC=0.932 for VS
CC=0.934 for MS
Non- and contrast-enhanced CT2985CNNAccurate CACS measurement as compared with manual methodCC=0.99Lee et al64
Sentitivity 93.3%
ECG-gated nonenhanced CCTA783CNN with 3D U-Net architectureThe DL method proposed yielded accuracy similar to those of other AI modelsAUC=0.951Gogin et al65
CCTA250Supervised CNNAn automatic CNN CACS method allowed fast and accurate coronary calcium detectionAUC=0.83Wolterink et al66
CACS CT7240Consecutive CNNsA DL CACS algorithm for quantification of coronary and thoracic aorta calcium was robust between different CT protocolsICC=0.79–0.97 for CACSvan Velzen et al67
Chest CT
PET attenuation correction CTICC=0.66–0.98 for TAC
Radiation therapy planning CT
Noncontrast chest CT5973CNNA convolutional neural network can directly regress the AS from the image of the heart without the need for prior segmentationCC=0.93Cano-Espinosa et al68
Low-dose noncontrast enhanced CT160Dilated convolutional layers; CNNsAlgorithm allowed to exclude negative scans for CACSDC=0.84van den Oever et al69
CCTA194RNN with LSTMHigh diagnostic accuracy for the detection of calcified coronary plaquesSensitivity=92.1%Fischer et al70
Specificity=88.9%
Accuracy=90.3%
CCTA78XGBoostThe algorithm was superior to the conventional cut-off method for coronary plaque characterizationAUC=0.92Masuda et al71
Noncontrast CCT2068DLEAT volume and attenuation quantified by automated DL algorithms provided additional prognostic value for prediction of MACE in asymptomatic subjectsHR=1.35 for EAT volumeEisenberg et al72
HR=0.83 for EAT attenuation
Noncontrast CCT250ConvNetCorrelation between fully automated quantification of EAT and thoracic adipose tissue in asymptomatic individuals and expert manual quantificationDC=0.82Commandeur et al73
CCTA220Unsupervised hierarchical clusteringAbility to discriminate high-risk plaque features and increased FAI as independent predictors of MACEsHoshino et al74
CCTA1069XGBoostML risk score derived from CCTA parameters and circulating biomarkers significantly predicted MI and cardiac death in long-term follow upAUC=0.81Tamarappoo et al75
CCTA203Deep stacking generalization framework and multiple levels of GBM classifiersA ML-based proteomic targeted radiogenomic approach was used to predict high-risk plaquesAUC=0.83Bom et al76
CCTA1931Random forest algorithmA validated ML-derived radiotranscriptomic approach was able to better predict MACEs as compared with traditional risk classificationHR=1.12Oikonomou et al77
PET/CT830LogitBoostML integration of CCTA and clinical data improved the identification of myocardial ischemiaAUC=0.85Benjamins et al78
FFR-CCTA87DNNDevelopment of an AI model with high diagnostic accuracy to detect cardiac FFR as compared with invasive methodAccuracy=83%Itu et al79
FFR-CCTA159Multilayer neural networkAI-based CT FFR was a better predictor for coronary revascularization and MACE than triple-rule-out CCTAMartin et al80
CCTA1052CNN, cGANA 3D DL model performed fully automatic estimation of abnormal FFRAUC=0.78Kumamaru et al81
CCTA351DNNA ML CT-FFR based model allowed to reclassify hemodynamically nonsignificant stenosisAccuracy=85%Coenen et al82
CCTA351DNNML-based CT-FFR showed a superior diagnostic performance for the detection of lesion-specific ischemia without sex differencesAUC=0.83Baumann et al83
CCTA33ML softwareA CT-FFR AI model showed a significant correlation with traditional model for the detection of plaque morphology and hemodynamically significant stenosisAUC=0.90Baumann et al84
CCTA85ML softwareFFR-ML algorithm performed equally to detect lesion-specific ischemia with the classical method, while showed a best performance for the identification of flow-limiting stenosisAUC=0.89Tesche et al85
CCTA84DLThe addition of ML-based CT-FFR analysis to CCTA derived plaque markers showed an incremental discriminatory power for the detection of stenosis degree as compared with CCTA assessment aloneAUC=0.93von Knebel Doeberitz et al86
CCTA314DNNAI-based CT-FFR algorithm showed a superior diagnostic value over conventional CCTA analysis in vessels with high and low/intermediate ASAUC=0.71 for CACS≥400Tesche et al87
CCTA296DLDL-FFRCT reduced the need for ICA with a low MACE rate in 2-y follow-upLiu et al88
CCTA115DNNML-based FFRCT was accurate for stent evaluation as compared with invasive FFR. Age and ΔFFRCT/length were predictors of MACE in 2-y follow-upAccuracy=0.85 for ISRTang et al89
AUC=0.787 to predict MACEs

2D indicates two dimensional; 3D, three dimensional; AI, artificial intelligence; AS, Agatson score; AUC, area under curve; CACS, coronary artery calcium score; CAD-RADS, Coronary Artery Disease-Reporting and Data System score; CC, correlation coefficient; CCTA, cardiac computed tomography angiography; cGAN, conditional generative adversarial network; CHD, coronary heart disease; CNN, convolutional neural network; CT, computed tomography; DC, dice coefficient score; DNN, deep neural network; DL, deep learning; EAT, epicardial adipose tissue; FAI, fat attenuation index; FFR, fractional flow reserve; GBM, gradient boosting machine; GUI, graphical user interface; HR, hazard ratio; ICA, invasive coronary angiography; ICC, intraclass correlation coefficient; ISR, in-stent restenosis; LSTM, long short-term memory; LV, left ventricle; LVM, left ventricular mass; MACE, major adverse cardiac event; ML, machine learning; MPV, mosaic projection view; MS, mass score; PET, positron emission tomography; RNN, recurrent neural network; TAC, thoracic aorta calcification; TL, transfer learning; VGG, visual geometry group; and VS, volume score.

AI-Based CMR Applications

Myocardial Tissue Characterization

An automated analysis method based on a fully convolutional neural network was trained and evaluated on a large-scale dataset of 4875 subjects from the UK Biobank. The performance of the method has been evaluated using a number of technical metrics and clinical parameters such as LV end-diastolic volume, LV end-systolic volume, LVM, RV end-diastolic volume, and RV end-systolic volume, providing results comparable to human interobserver variability in segmenting the LV and RV on short-axis images and the left atrium and right atrium on long-axis CMR images.90

Three CNNs with the U-NET architecture were tested on 3 training datasets of cine MR images with increasing variability levels. A CNN, based on dataset heterogeneity, was able to accurately quantify LV function parameters as compared with manual analysis with a correlation of 0.98.91 Ruijsink et al developed a DL-based algorithm framework for cardiac function analysis including quality controls in each phase of the pipeline (image preanalysis, analytical phase, post analysis phase aimed to identify erroneous results). The automated analysis showed a high correlation with manual analysis according to left and right ventricular volumes, strain, and filling and ejection parameters.92

The application of AI was also explored in LGE CMR for cardiac tissue characterization. Zabihollahy et al, applied to LGE CMR images a 3D CNN algorithm with the aim to discriminate ventricular tissue scars from healthy tissues. The algorithm was compared with the results obtained with manual segmentation obtaining a dice similarity coefficient of 0.94.93 In addition, Moccia et al94 reported a DCS of 0.88 by using a fully convolutional neural network to segment LV scar tissue from LGE images of patients with CHD in comparison with traditional segmentation method.

In an interesting study by Zangh et al applied in patients with MI an automatic DL framework by exploiting nonenhanced cine CMR. Data extrapolated from AI analysis of nonenhanced cine CMR versus LGE confirmed the presence, position, and transmurality/size of the infarcted area (sensitivity, specificity, and AUC of 89.8%, 99.1%, and 0.94, respectively) with the advantage to avoid gadolinium injection.95

A DL approach on a cohort of 32 239 from the UK Biobank showed that LVM-AI derived was correlated with LVM-derived CMR and appeared more accurate for LVH diagnosis. Furthermore, LVH predicted by LVM-AI was significant predictive of adverse cardiovascular events such as MI (HR=1.80).96

Myocardial Perfusion

Kim et al97 trained a U-Net segmentation model and Monte Carlo dropout sampling for the segmentation of endocardial or epicardial dynamic contrast enhancement images obtaining a dice similarity coefficient comparable to the semiautomatic method.

In stress perfusion CMR, a CNN approach from the automatic segmentation of LV cavity and myocardium was applied and perfusion maps, myocardial blood flow and myocardial perfusion reserve parameters provided outputs comparable to manual methodology.98 In addition, in patients with known or suspected CHD reduced myocardial blood flow and myocardial perfusion reserve resulted as independent predictors of death and MACE99 (Table 3).

Table 3. CMR-Based AI Applications in CHD

Imaging protocolParticipantsAI applicationResultsPerformanceReference
Cine CMR4875FCNNAn automated segmentation method achieves a high performance as compared with manual method for the LV and RV on short-axis images and for LA and RA on long-axis imagesDSC=0.94 for LV cavityBai et al90
DSC=0.88 for LV myocardium
DSC=0.90 for RV cavity
DSC=0.93 for LA cavity (2Ch v)
DSC=0.95 for LA (4Ch v)
DSC=0.96 for RA cavity (4Ch v)
Cine CMR8963 CNNsFully automated quantification of LV function from short-axis images showed high correlation with the parameters obtained by experienced observersCC=0.98Tao et al91
Cine CMR2029DLAutomated analysis highly correlated with manual segmentation for LV and RV volumes, strain, filling and ejection ratesSensitivity=95%Ruijsink et al92
CC range=0.89–0.95
LGE CMR343D CNNDevelopment of a semiautomated methodology for 3D segmentation of LV and myocardial scar in comparison to other segmentation methodologiesDSC=0.93Zabihollahy et al93
LGE CMR30FCNNIntroduction of 2 segmentation protocols able to detect scar tissues with higher performance when limiting the search area to the myocardial regionDSC=0.88Moccia et al94
Noncontrast enhanced cine CMR299FCDNA DL framework able to detect presence, position, size, and transmurality of chronic MIAUC=0.94Zhang et al95
CMR32 239One dimensional CNNLVM-AI was correlated with LVM-derived CMR and appeared more accurate for LVH diagnosis. LVH predicted by LVM-AI was associated with cardiovascular events such as MICC=0.79Khurshid et al96
HR= 1.80 for incidence of MI
DCE perfusion CMR35Trained segmentation CNN model of cine dataThe automatic AI model applied to perfusion data showed a robust myocardial segmentation as compared with semiautomatic methodsDSC=0.806Kim et al97
Adenosine stress and rest perfusion CMR1139CNNThe CNN performed similarly to manual segmentation and flow measuresDSC=0.93Xue H et al98
Cine CMR1049CNNReduction of MBF and MPR, quantified by automated approach, resulted strong and independent predictors of death and MACEsHR=1.93/2.14 for MBFKnott et al99
Stress and rest perfusion CMR
HR=2.45/1.74 for MPR
LGE CMR

3D indicates three dimensional; AI, artificial intelligence; AUC, area under the curve; CC, correlation coefficient; CHD, coronary heart disease; Ch v, chamber view; CMR, cardiac magnetic resonance; CNN, convolutional neural network; DCE, dynamic contrast enhancement; DL, deep learning; DSC, dice similarity coefficient; FCDN, fully connected discriminative network; FCNN, fully convolutional neural network; HR, hazard ratio; LA, left atrium; LGE, late gadolinium enhancement; LV, left ventricle; LVH, left ventricle hypertrophy; LVM, left ventricular mass; MACE, major adverse cardiac event; MBF, myocardial blood flow; MI, myocardial infarction; MPR, myocardial perfusion reserve; RA, right atrium; and RV, right ventricle.

Cetin et al identified specific radiomic signatures from CMR images of a large cohort of UK Bank participants. ML algorithms (support vector machine, random forest, and logistic regression) showed that specific radiomic signatures were able to differentiate healthy subjects from individuals with specific risk factors (hypertension, diabetes, smoking, and high cholesterol). In addition, with respect to conventional CMR indices, radiomic features provided better discrimination between controls and at-risk subjects49

Conclusions

Radiomics, radiogenomics, and AI approaches emerged as promising applications in CHD. Indeed, radiomics demonstrated to allow an accurate quantitative analysis of plaque morphology and composition by CCTA and myocardial tissue by CMR. In addition, AI has found several promising findings reducing the time of reporting, and improving the task of precision medicine. The multilayer approach of ML could unravel hidden information in heterogeneous datasets and fill the gap between disease pathogenesis, genotypes, and phenotypes. Despite these encouraging perspectives in the identification and characterization of CHD and patient outcome prediction, there are unsolved issues for the translation of radiomics, radiogenomics, and AI-based tools into clinical practice (Figure 2). Concerning radiomics, the extracted features are affected by preprocessing steps before the calculation of parameters, thus showing the need for standardized pipelines for quality check, analysis, and validation. In addition, an important drawback is the biological and imaging database heterogeneity or low resolution images in datasets, which limits the applicability of ML algorithms. In addition, for genomic tool analysis, an important issue is represented by the unstandardization of tools and modifications in their versions and outcome format. Since the number and types of radiomic/genomics and clinical variables in ML algorithms are arbitrarily decided by the operator, a question arises regarding the optimal standard and validated procedures, since an excessive amount of parameters could provide an overestimation analysis of results. In this regard, the characteristics of the included studies, such as patient cohorts, study aim and setting, imaging protocols and analysis methods, were highly variable across studies, preventing us from performing a meta-analysis.

In addition, the lack of transparency of AI tools, which are often not easy intelligible, and of objectivity makes the AI outputs susceptible to inaccuracies and bias, needing necessary the development of AI algorithms and software interface user friendly for clinicians and researchers.

Furthermore, ethical surveillance practices for data decoding, anonymization, sharing and collection and privacy policies, in particular, in case of sensitive data are necessary.

In light of this, several efforts are still needed to validate and standardize algorithms for CHD patient routine according to good medical practice.

ARTICLE INFORMATION

Supplemental Material

Tables S1–S4

Nonstandard Abbreviations and Acronyms

AI

artificial intelligence

AUC

area under the curve

CACS

coronary calcium score

CCTA

cardiac computed tomography angiography

CHD

coronary heart disease

CMR

cardiac magnetic resonance

CNN

convolutional neural network

DL

deep learning

EAT

epicardial adipose tissue

FFR

fractional flow reserve

GDF-15

growth differentiation factor-15

LGE

late gadolinium enhancement

LV

left ventricle

LVEF

left ventricular ejection fraction

LVM

left ventricular mass

MACE

major adverse cardiovascular event

MI

myocardial infarction

ML

machine learning

MMP-9

matrix metallopeptidase 9

PIGR

polymeric immunoglobulin receptor

STEMI

ST-segment elevation myocardial infarction

Disclosures None.

Footnotes

Supplemental Material is available at https://www.ahajournals.org/doi/suppl/10.1161/CIRCIMAGING.121.013025.

For Sources of Funding and Disclosures, see page 1143.

Correspondence to: Teresa Infante, Biol.D, Department of Advanced Medical and Surgical Sciences, University of Campania “Luigi Vanvitelli”, Piazza Miraglia, 2, 80138 Naples, Italy. Email

References

  • 1. Arnett DK, Blumenthal RS, Albert MA, Buroker AB, Goldberger ZD, Hahn EJ, Himmelfarb CD, Khera A, Lloyd-Jones D, McEvoy JWet al. 2019 ACC/AHA guideline on the primary prevention of cardiovascular disease: executive summary: a report of the American College of Cardiology/American Heart Association Task Force on Clinical Practice Guidelines.Circulation. 2019; 140:e563–e595. doi: 10.1161/CIR.0000000000000677LinkGoogle Scholar
  • 2. Napoli C, Lerman LO, de Nigris F, Gossl M, Balestrieri ML, Lerman A. Rethinking primary prevention of atherosclerosis-related diseases.Circulation. 2006; 114:2517–2527. doi: 10.1161/CIRCULATIONAHA.105.570358LinkGoogle Scholar
  • 3. Napoli C, Crudele V, Soricelli A, Al-Omran M, Vitale N, Infante T, Mancini FP. Primary prevention of atherosclerosis: a clinical challenge for the reversal of epigenetic mechanisms?Circulation. 2012; 125:2363–2373. doi: 10.1161/CIRCULATIONAHA.111.085787LinkGoogle Scholar
  • 4. Piepoli MF, Abreu A, Albus C, Ambrosetti M, Brotons C, Catapano AL, Corra U, Cosyns B, Deaton C, Graham Iet al. Update on cardiovascular prevention in clinical practice: a position paper of the European Association of Preventive Cardiology of the European Society of Cardiology.Eur J Prev Cardiol. 2020; 27:181–205. doi: 10.1177/2047487319893035CrossrefMedlineGoogle Scholar
  • 5. Schiano C, Vietri MT, Grimaldi V, Picascia A, De Pascale MR, Napoli C. Epigenetic-related therapeutic challenges in cardiovascular disease.Trends Pharmacol Sci. 2015; 36:226–235. doi: 10.1016/j.tips.2015.02.005CrossrefMedlineGoogle Scholar
  • 6. Infante T, Forte E, Schiano C, Cavaliere C, Tedeschi C, Soricelli A, Salvatore M, Napoli C. An integrated approach to coronary heart disease diagnosis and clinical management.Am J Transl Res. 2017; 9:3148–3166.MedlineGoogle Scholar
  • 7. Infante T, Del Viscovo L, De Rimini ML, Padula S, Caso P, Napoli C. Network medicine: a clinical approach for precision medicine and personalized therapy in coronary heart disease.J Atheroscler Thromb. 2020; 27:279–302. doi: 10.5551/jat.52407CrossrefMedlineGoogle Scholar
  • 8. Infante T, Forte E, Punzo B, Cademartiri F, Cavaliere C, Soricelli A, Salvatore M, Napoli C. Correlation of Circulating miR-765, miR-93-5p, and miR-433-3p to obstructive coronary heart disease evaluated by cardiac computed tomography.Am J Cardiol. 2019; 124:176–182. doi: 10.1016/j.amjcard.2019.04.016CrossrefMedlineGoogle Scholar
  • 9. Newby DE, Adamson PD, Berry C, Boon NA, Dweck MR, Flather M, Forbes J, Hunter A, Lewis S, MacLean Set al; SCOT-HEART Investigators. Coronary CT angiography and 5-year risk of myocardial infarction.N Engl J Med. 2018; 379:924–933. doi: 10.1056/NEJMoa1805971CrossrefMedlineGoogle Scholar
  • 10. Forte E, Aiello M, Inglese M, Infante T, Soricelli A, Tedeschi C, Salvatore M, Cavaliere C. Coronary artery aneurysms detected by computed tomography coronary angiography.Eur Heart J Cardiovasc Imaging. 2017; 18:1229–1235. doi: 10.1093/ehjci/jew218CrossrefMedlineGoogle Scholar
  • 11. Ahmadi A, Leipsic J, Øvrehus KA, Gaur S, Bagiella E, Ko B, Dey D, LaRocca G, Jensen JM, Bøtker HEet al. Lesion-specific and vessel-related determinants of fractional flow reserve beyond coronary artery stenosis.JACC Cardiovasc Imaging. 2018; 11:521–530. doi: 10.1016/j.jcmg.2017.11.020CrossrefMedlineGoogle Scholar
  • 12. Feuchtner G, Kerber J, Burghard P, Dichtl W, Friedrich G, Bonaros N, Plank F. The high-risk criteria low-attenuation plaque <60 HU and the napkin-ring sign are the most powerful predictors of MACE: a long-term follow-up study.Eur Heart J Cardiovasc Imaging. 2017; 18:772–779. doi: 10.1093/ehjci/jew167CrossrefMedlineGoogle Scholar
  • 13. Hajhosseiny R, Bustin A, Munoz C, Rashid I, Cruz G, Manning WJ, Prieto C, Botnar RM. Coronary magnetic resonance angiography: technical innovations leading us to the promised land?JACC Cardiovasc Imaging. 2020; 13:2653–2672. doi: 10.1016/j.jcmg.2020.01.006CrossrefMedlineGoogle Scholar
  • 14. El Aidi H, Adams A, Moons KG, Den Ruijter HM, Mali WP, Doevendans PA, Nagel E, Schalla S, Bots ML, Leiner T. Cardiac magnetic resonance imaging findings and the risk of cardiovascular events in patients with recent myocardial infarction or suspected or known coronary artery disease: a systematic review of prognostic studies.J Am Coll Cardiol. 2014; 63:1031–1045. doi: 10.1016/j.jacc.2013.11.048CrossrefMedlineGoogle Scholar
  • 15. Gronda E, Vanoli E, Sacchi S, Grassi G, Ambrosio G, Napoli C. Risk of heart failure progression in patients with reduced ejection fraction: mechanisms and therapeutic options.Heart Fail Rev. 2020; 25:295–303. doi: 10.1007/s10741-019-09823-zCrossrefMedlineGoogle Scholar
  • 16. Xu P, Xue Y, Schoepf UJ, Varga-Szemes A, Griffith J, Yacoub B, Zhou F, Zhou C, Yang Y, Xing Wet al. Radiomics: the next frontier of cardiac computed tomography.Circ Cardiovasc Imaging. 2021; 14:e011747. doi: 10.1161/CIRCIMAGING.120.011747LinkGoogle Scholar
  • 17. Schiano C, Benincasa G, Infante T, Franzese M, Castaldo R, Fiorito C, Mansueto G, Grimaldi V, Della Valle G, Fatone Get al. Integrated analysis of DNA methylation profile of HLA-G gene and imaging in coronary heart disease: pilot study.PLoS One. 2020; 15:e0236951. doi: 10.1371/journal.pone.0236951CrossrefMedlineGoogle Scholar
  • 18. Infante T, Forte E, Schiano C, Punzo B, Cademartiri F, Cavaliere C, Salvatore M, Napoli C. Evidence of association of circulating epigenetic-sensitive biomarkers with suspected coronary heart disease evaluated by Cardiac Computed Tomography.PLoS One. 2019; 14:e0210909. doi: 10.1371/journal.pone.0210909CrossrefMedlineGoogle Scholar
  • 19. Infante T, Forte E, Aiello M, Salvatore M, Cavaliere C. In vvo and in vitro analysis in coronary artery disease related to type 2 diabetes.Front Endocrinol (Lausanne). 2017; 8:209. doi: 10.3389/fendo.2017.00209CrossrefMedlineGoogle Scholar
  • 20. Aiello M, Cavaliere C, D’Albore A, Salvatore M. The challenges of diagnostic imaging in the era of big data.J Clin Med. 2019; 8:E316. doi: 10.3390/jcm8030316CrossrefMedlineGoogle Scholar
  • 21. Schiano C, Benincasa G, Franzese M, Della Mura N, Pane K, Salvatore M, Napoli C. Epigenetic-sensitive pathways in personalized therapy of major cardiovascular diseases.Pharmacol Ther. 2020; 210:107514. doi: 10.1016/j.pharmthera.2020.107514CrossrefMedlineGoogle Scholar
  • 22. Benincasa G, Marfella R, Della Mura N, Schiano C, Napoli C. Strengths and opportunities of network medicine in cardiovascular diseases.Circ J. 2020; 84:144–152. doi: 10.1253/circj.CJ-19-0879CrossrefMedlineGoogle Scholar
  • 23. Incoronato M, Aiello M, Infante T, Cavaliere C, Grimaldi AM, Mirabelli P, Monti S, Salvatore M. Radiogenomic analysis of oncological data: a technical survey.Int J Mol Sci. 2017; 18:E805. doi: 10.3390/ijms18040805CrossrefMedlineGoogle Scholar
  • 24. Murgia A, Balestrieri A, Crivelli P, Suri JS, Conti M, Cademartiri F, Saba L. Cardiac computed tomography radiomics: an emerging tool for the non-invasive assessment of coronary atherosclerosis.Cardiovasc Diagn Ther. 2020; 10:2005–2017. doi: 10.21037/cdt-20-156CrossrefMedlineGoogle Scholar
  • 25. Hassani C, Saremi F, Varghese BA, Duddalwar V. Myocardial radiomics in cardiac MRI.AJR Am J Roentgenol. 2020; 214:536–545. doi: 10.2214/AJR.19.21986CrossrefMedlineGoogle Scholar
  • 26. Beam AL, Kohane IS. Big data and machine learning in health care.JAMA. 2018; 319:1317–1318. doi: 10.1001/jama.2017.18391CrossrefMedlineGoogle Scholar
  • 27. Infante T, Francone M, De Rimini ML, Cavaliere C, Canonico R, Catalano C, Napoli C. Machine learning and network medicine: a novel approach for precision medicine and personalized therapy in cardiomyopathies.J Cardiovasc Med (Hagerstown). 2021; 22:429–440. doi: 10.2459/JCM.0000000000001103CrossrefMedlineGoogle Scholar
  • 28. Lin A, Kolossváry M, Motwani M, Išgum I, Maurovich-Horvat P, Slomka PJ, Dey D. Artificial intelligence in cardiovascular imaging for risk stratification in coronary artery disease.Radiol Cardiothorac Imaging. 2021; 3:e200512. doi: 10.1148/ryct.2021200512CrossrefMedlineGoogle Scholar
  • 29. Maurovich-Horvat P. Current trends in the use of machine learning for diagnostics and/or risk stratification in cardiovascular disease.Cardiovasc Res. 2021:cvab059. doi: 10.1093/cvr/cvab059CrossrefGoogle Scholar
  • 30. Currie G, Hawk KE, Rohren E, Vial A, Klein R. Machine learning and deep learning in medical imaging: intelligent imaging.J Med Imaging Radiat Sci. 2019; 50:477–487. doi: 10.1016/j.jmir.2019.09.005CrossrefMedlineGoogle Scholar
  • 31. Joshi A, Rienks M, Theofilatos K, Mayr M. Systems biology in cardiovascular disease: a multiomics approach.Nat Rev Cardiol. 2021; 18:313–330. doi: 10.1038/s41569-020-00477-1CrossrefMedlineGoogle Scholar
  • 32. Visco V, Ferruzzi GJ, Nicastro F, Virtuoso N, Carrizzo A, Galasso G, Vecchione C, Ciccarelli M. Artificial intelligence as a business partner in cardiovascular precision medicine: an emerging approach for disease detection and treatment optimization.Curr Med Chem. 2021; 28:6569–6590. doi: 10.2174/0929867328666201218122633CrossrefMedlineGoogle Scholar
  • 33. Majnarić LT, Babič F, O’Sullivan S, Holzinger A. AI and big data in healthcare: towards a more comprehensive research framework for multimorbidity.J Clin Med. 2021; 10:766. doi: 10.3390/jcm10040766CrossrefMedlineGoogle Scholar
  • 34. Moher D, Shamseer L, Clarke M, Ghersi D, Liberati A, Petticrew M, Shekelle P, Stewart LA; PRISMA-P Group. Preferred reporting items for systematic review and meta-analysis protocols (PRISMA-P) 2015 statement.Syst Rev. 2015; 4:1. doi: 10.1186/2046-4053-4-1CrossrefMedlineGoogle Scholar
  • 35. Gillies RJ, Kinahan PE, Hricak H. Radiomics: images are more than pictures, they are data.Radiology. 2016; 278:563–577. doi: 10.1148/radiol.2015151169CrossrefMedlineGoogle Scholar
  • 36. Castaldo R, Cavaliere C, Soricelli A, Salvatore M, Pecchia L, Franzese M. Radiomic and genomic machine learning method performance for prostate cancer diagnosis: systematic literature review.J Med Internet Res. 2021; 23:e22394. doi: 10.2196/22394CrossrefMedlineGoogle Scholar
  • 37. Koçak B, Durmaz EŞ, Ateş E, Kiliçkesmez Ö. Radiomics with artificial intelligence: a practical guide for beginners.Diagn Interv Radiol. 2019; 25:485–495. doi: 10.5152/dir.2019.19321CrossrefMedlineGoogle Scholar
  • 38. Hu W, Wu X, Dong D, Cui LB, Jiang M, Zhang J, Wang Y, Wang X, Gao L, Tian J, Cao F. Novel radiomics features from CCTA images for the functional evaluation of significant ischaemic lesions based on the coronary fractional flow reserve score.Int J Cardiovasc Imaging. 2020; 36:2039–2050. doi: 10.1007/s10554-020-01896-4CrossrefMedlineGoogle Scholar
  • 39. Kolossváry M, Park J, Bang JI, Zhang J, Lee JM, Paeng JC, Merkely B, Narula J, Kubo T, Akasaka Tet al. Identification of invasive and radionuclide imaging markers of coronary plaque vulnerability using radiomic analysis of coronary computed tomography angiography.Eur Heart J Cardiovasc Imaging. 2019; 20:1250–1258. doi: 10.1093/ehjci/jez033CrossrefMedlineGoogle Scholar
  • 40. Kolossváry M, Karády J, Szilveszter B, Kitslaar P, Hoffmann U, Merkely B, Maurovich-Horvat P. Radiomic features are superior to conventional quantitative computed tomographic metrics to identify coronary plaques with napkin-ring sign.Circ Cardiovasc Imaging. 2017; 10:e006843. doi: 10.1161/CIRCIMAGING.117.006843LinkGoogle Scholar
  • 41. Lin A, Kolossváry M, Yuvaraj J, Cadet S, McElhinney PA, Jiang C, Nerlekar N, Nicholls SJ, Slomka PJ, Maurovich-Horvat Pet al. Myocardial infarction associates with a distinct pericoronary adipose tissue radiomic phenotype: a prospective case-control study.JACC Cardiovasc Imaging. 2020; 13:2371–2383. doi: 10.1016/j.jcmg.2020.06.033CrossrefMedlineGoogle Scholar
  • 42. Mannil M, von Spiczak J, Manka R, Alkadhi H. Texture analysis and machine learning for detecting myocardial infarction in noncontrast low-dose computed tomography: unveiling the invisible.Invest Radiol. 2018; 53:338–343. doi: 10.1097/RLI.0000000000000448CrossrefMedlineGoogle Scholar
  • 43. Bulluck H, Dharmakumar R, Arai AE, Berry C, Hausenloy DJ. Cardiovascular magnetic resonance in acute ST-segment-elevation myocardial infarction: recent advances, controversies, and future directions.Circulation. 2018; 137:1949–1964. doi: 10.1161/CIRCULATIONAHA.117.030693LinkGoogle Scholar
  • 44. Baessler B, Mannil M, Oebel S, Maintz D, Alkadhi H, Manka R. Subacute and chronic left ventricular myocardial scar: accuracy of texture analysis on nonenhanced cine MR images.Radiology. 2018; 286:103–112. doi: 10.1148/radiol.2017170213CrossrefMedlineGoogle Scholar
  • 45. Larroza A, López-Lereu MP, Monmeneu JV, Gavara J, Chorro FJ, Bodí V, Moratal D. Texture analysis of cardiac cine magnetic resonance imaging to detect nonviable segments in patients with chronic myocardial infarction.Med Phys. 2018; 45:1471–1480. doi: 10.1002/mp.12783CrossrefMedlineGoogle Scholar
  • 46. Larroza A, Materka A, López-Lereu MP, Monmeneu JV, Bodí V, Moratal D. Differentiation between acute and chronic myocardial infarction by means of texture analysis of late gadolinium enhancement and cine cardiac magnetic resonance imaging.Eur J Radiol. 2017; 92:78–83. doi: 10.1016/j.ejrad.2017.04.024CrossrefMedlineGoogle Scholar
  • 47. Chen BH, An DA, He J, Wu CW, Yue T, Wu R, Shi RY, Eteer K, Joseph B, Hu Jet al. Myocardial extracellular volume fraction radiomics analysis for differentiation of reversible versus irreversible myocardial damage and prediction of left ventricular adverse remodeling after ST-elevation myocardial infarction.Eur Radiol. 2021; 31:504–514. doi: 10.1007/s00330-020-07117-9CrossrefMedlineGoogle Scholar
  • 48. Ma Q, Ma Y, Yu T, Sun Z, Hou Y. Radiomics of non-contrast-enhanced T1 mapping: diagnostic and predictive performance for myocardial injury in acute ST-segment-elevation myocardial infarction.Korean J Radiol. 2021; 22:535–546. doi: 10.3348/kjr.2019.0969CrossrefMedlineGoogle Scholar
  • 49. Cetin I, Raisi-Estabragh Z, Petersen SE, Napel S, Piechnik SK, Neubauer S, Gonzalez Ballester MA, Camara O, Lekadir K. Radiomics signatures of cardiovascular risk factors in cardiac MRI: results from the UK biobank.Front Cardiovasc Med. 2020; 7:591368. doi: 10.3389/fcvm.2020.591368CrossrefMedlineGoogle Scholar
  • 50. Ranka S, Reddy M, Noheria A. Artificial intelligence in cardiovascular medicine.Curr Opin Cardiol. 2021; 36:26–35. doi: 10.1097/HCO.0000000000000812CrossrefMedlineGoogle Scholar
  • 51. White RD, Erdal BS, Demirer M, Gupta V, Bigelow MT, Dikici E, Candemir S, Galizia MS, Carpenter JL, O’Donnell TPet al. Artificial intelligence to assist in exclusion of coronary atherosclerosis during CCTA evaluation of chest pain in the emergency department: preparing an application for real-world use.J Digit Imaging. 2021; 34:554–571. doi: 10.1007/s10278-021-00441-6CrossrefMedlineGoogle Scholar
  • 52. Han D, Liu J, Sun Z, Cui Y, He Y, Yang Z. Deep learning analysis in coronary computed tomographic angiography imaging for the assessment of patients with coronary artery stenosis.Comput Methods Programs Biomed. 2020; 196:105651. doi: 10.1016/j.cmpb.2020.105651CrossrefMedlineGoogle Scholar
  • 53. Zreik M, van Hamersvelt RW, Wolterink JM, Leiner T, Viergever MA, Isgum I. A recurrent CNN for automatic detection and classification of coronary artery plaque and stenosis in coronary CT angiography.IEEE Trans Med Imaging. 2019; 38:1588–1598. doi: 10.1109/TMI.2018.2883807CrossrefMedlineGoogle Scholar
  • 54. Podgorsak AR, Sommer KN, Reddy A, Iyer V, Wilson MF, Rybicki FJ, Mitsouras D, Sharma U, Fujimoto S, Kumamaru KKet al. Initial evaluation of a convolutional neural network used for noninvasive assessment of coronary artery disease severity from coronary computed tomography angiography data.Med Phys. 2020; 47:3996–4004. doi: 10.1002/mp.14339CrossrefMedlineGoogle Scholar
  • 55. Zreik M, van Hamersvelt RW, Khalili N, Wolterink JM, Voskuil M, Viergever MA, Leiner T, Isgum I. Deep learning analysis of coronary arteries in cardiac CT angiography for detection of patients requiring invasive coronary angiography.IEEE Trans Med Imaging. 2020; 39:1545–1557. doi: 10.1109/TMI.2019.2953054CrossrefMedlineGoogle Scholar
  • 56. van Hamersvelt RW, Zreik M, Voskuil M, Viergever MA, Išgum I, Leiner T. Deep learning analysis of left ventricular myocardium in CT angiographic intermediate-degree coronary stenosis improves the diagnostic accuracy for identification of functionally significant stenosis.Eur Radiol. 2019; 29:2350–2359. doi: 10.1007/s00330-018-5822-3CrossrefMedlineGoogle Scholar
  • 57. Zreik M, Lessmann N, van Hamersvelt RW, Wolterink JM, Voskuil M, Viergever MA, Leiner T, Išgum I. Deep learning analysis of the myocardium in coronary CT angiography for identification of patients with functionally significant coronary artery stenosis.Med Image Anal. 2018; 44:72–85. doi: 10.1016/j.media.2017.11.008CrossrefMedlineGoogle Scholar
  • 58. Muscogiuri G, Chiesa M, Trotta M, Gatti M, Palmisano V, Dell’Aversana S, Baessato F, Cavaliere A, Cicala G, Loffreno Aet al. Performance of a deep learning algorithm for the evaluation of CAD-RADS classification with CCTA.Atherosclerosis. 2020; 294:25–32. doi: 10.1016/j.atherosclerosis.2019.12.001CrossrefMedlineGoogle Scholar
  • 59. Huang Z, Xiao J, Xie Y, Hu Y, Zhang S, Li X, Wang Z, Li Z, Wang X. The correlation of deep learning-based CAD-RADS evaluated by coronary computed tomography angiography with breast arterial calcification on mammography.Sci Rep. 2020; 10:11532. doi: 10.1038/s41598-020-68378-4CrossrefMedlineGoogle Scholar
  • 60. Choi AD, Marques H, Kumar V, Griffin WF, Rahban H, Karlsberg RP, Zeman RK, Katz RJ, Earls JP. CT evaluation by artificial intelligence for atherosclerosis, stenosis and vascular morphology (CLARIFY): a multi-center, international study.J Cardiovasc Comput Tomogr. 2021; 15:470–476. S1934-5925(21)00081-2. doi: 10.1016/j.jcct.2021.05.004CrossrefMedlineGoogle Scholar
  • 61. Głowacki J, Krysiński M, Czaja-Ziółkowska M, Wasilewski J. Machine learning-based algorithm enables the exclusion of obstructive coronary artery disease in the patients who underwent coronary artery calcium scoring.Acad Radiol. 2020; 27:1416–1421. doi: 10.1016/j.acra.2019.11.016CrossrefMedlineGoogle Scholar
  • 62. Wang W, Wang H, Chen Q, Zhou Z, Wang R, Wang H, Zhang N, Chen Y, Sun Z, Xu L. Coronary artery calcium score quantification using a deep-learning algorithm.Clin Radiol. 2020; 75:237.e11–237.e16. doi: 10.1016/j.crad.2019.10.012CrossrefMedlineGoogle Scholar
  • 63. Sandstedt M, Henriksson L, Janzon M, Nyberg G, Engvall J, De Geer J, Alfredsson J, Persson A. Evaluation of an AI-based, automatic coronary artery calcium scoring software.Eur Radiol. 2020; 30:1671–1678. doi: 10.1007/s00330-019-06489-xCrossrefMedlineGoogle Scholar
  • 64. Lee JG, Kim H, Kang H, Koo HJ, Kang JW, Kim YH, Yang DH. Fully automatic coronary calcium score software empowered by artificial intelligence technology: validation study using three CT cohorts.Korean J Radiol. 2021; 22:1764–1776. doi: 10.3348/kjr.2021.0148CrossrefMedlineGoogle Scholar
  • 65. Gogin N, Viti M, Nicodème L, Ohana M, Talbot H, Gencer U, Mekukosokeng M, Caramella T, Diascorn Y, Airaud JYet al. Automatic coronary artery calcium scoring from unenhanced-ECG-gated CT using deep learning.Diagn Interv Imaging. 2021:S2211–5684(21)00118-2. doi: 10.1016/j.diii.2021.05.004CrossrefGoogle Scholar
  • 66. Wolterink JM, Leiner T, de Vos BD, van Hamersvelt RW, Viergever MA, Išgum I. Automatic coronary artery calcium scoring in cardiac CT angiography using paired convolutional neural networks.Med Image Anal. 2016; 34:123–136. doi: 10.1016/j.media.2016.04.004CrossrefMedlineGoogle Scholar
  • 67. van Velzen SGM, Lessmann N, Velthuis BK, Bank IEM, van den Bongard DHJG, Leiner T, de Jong PA, Veldhuis WB, Correa A, Terry JGet al. Deep learning for automatic calcium scoring in CT: validation using multiple cardiac CT and chest CT protocols.Radiology. 2020; 295:66–79. doi: 10.1148/radiol.2020191621CrossrefMedlineGoogle Scholar
  • 68. Cano-Espinosa C, González G, Washko GR, Cazorla M, Estépar RSJ. Automated agatston score computation in non-ECG gated CT scans using deep learning.Proc SPIE Int Soc Opt Eng. 2018; 10574:105742K. doi: 10.1117/12.2293681CrossrefMedlineGoogle Scholar
  • 69. van den Oever LB, Cornelissen L, Vonder M, Xia C, van Bolhuis JN, Vliegenthart R, Veldhuis RNJ, de Bock GH, Oudkerk M, van Ooijen PMA. Deep learning for automated exclusion of cardiac CT examinations negative for coronary artery calcium.Eur J Radiol. 2020; 129:109114. doi: 10.1016/j.ejrad.2020.109114CrossrefMedlineGoogle Scholar
  • 70. Fischer AM, Eid M, De Cecco CN, Gulsun MA, van Assen M, Nance JW, Sahbaee P, De Santis D, Bauer MJ, Jacobs BEet al. 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.J Thorac Imaging. 2020; 35(suppl 1):S49–S57. doi: 10.1097/RTI.0000000000000491CrossrefMedlineGoogle Scholar
  • 71. Masuda T, Nakaura T, Funama Y, Okimoto T, Sato T, Higaki T, Noda N, Imada N, Baba Y, Awai K. Machine-learning integration of CT histogram analysis to evaluate the composition of atherosclerotic plaques: validation with IB-IVUS.J Cardiovasc Comput Tomogr. 2019; 13:163–169. doi: 10.1016/j.jcct.2018.10.018CrossrefMedlineGoogle Scholar
  • 72. Eisenberg E, McElhinney PA, Commandeur F, Chen X, Cadet S, Goeller M, Razipour A, Gransar H, Cantu S, Miller RJHet al. Deep learning-based quantification of epicardial adipose tissue volume and attenuation predicts major adverse cardiovascular events in asymptomatic subjects.Circ Cardiovasc Imaging. 2020; 13:e009829. doi: 10.1161/CIRCIMAGING.119.009829LinkGoogle Scholar
  • 73. Commandeur F, Goeller M, Betancur J, Cadet S, Doris M, Chen X, Berman DS, Slomka PJ, Tamarappoo BK, Dey D. Deep learning for quantification of epicardial and thoracic adipose tissue from non-contrast CT.IEEE Trans Med Imaging. 2018; 37:1835–1846. doi: 10.1109/TMI.2018.2804799CrossrefMedlineGoogle Scholar
  • 74. Hoshino M, Zhang J, Sugiyama T, Yang S, Kanaji Y, Hamaya R, Yamaguchi M, Hada M, Misawa T, Usui Eet al. Prognostic value of pericoronary inflammation and unsupervised machine-learning-defined phenotypic clustering of CT angiographic findings.Int J Cardiol. 2021; 333:226–232. doi: 10.1016/j.ijcard.2021.03.019CrossrefMedlineGoogle Scholar
  • 75. Tamarappoo BK, Lin A, Commandeur F, McElhinney PA, Cadet S, Goeller M, Razipour A, Chen X, Gransar H, Cantu Set al. Machine learning integration of circulating and imaging biomarkers for explainable patient-specific prediction of cardiac events: a prospective study.Atherosclerosis. 2021; 318:76–82. doi: 10.1016/j.atherosclerosis.2020.11.008CrossrefMedlineGoogle Scholar
  • 76. Bom MJ, Levin E, Driessen RS, Danad I, Van Kuijk CC, van Rossum AC, Narula J, Min JK, Leipsic JA, Belo Pereira JPet al. Predictive value of targeted proteomics for coronary plaque morphology in patients with suspected coronary artery disease.EBioMedicine. 2019; 39:109–117. doi: 10.1016/j.ebiom.2018.12.033CrossrefMedlineGoogle Scholar
  • 77. Oikonomou EK, Williams MC, Kotanidis CP, Desai MY, Marwan M, Antonopoulos AS, Thomas KE, Thomas S, Akoumianakis I, Fan LMet al. A novel machine learning-derived radiotranscriptomic signature of perivascular fat improves cardiac risk prediction using coronary CT angiography.Eur Heart J. 2019; 40:3529–3543. doi: 10.1093/eurheartj/ehz592CrossrefMedlineGoogle Scholar
  • 78. Benjamins JW, Yeung MW, Maaniitty T, Saraste A, Klén R, van der Harst P, Knuuti J, Juarez-Orozco LE. Improving patient identification for advanced cardiac imaging through machine learning-integration of clinical and coronary CT angiography data.Int J Cardiol. 2021; 335:130–136. doi: 10.1016/j.ijcard.2021.04.009CrossrefMedlineGoogle Scholar
  • 79. Itu L, Rapaka S, Passerini T, Georgescu B, Schwemmer C, Schoebinger M, Flohr T, Sharma P, Comaniciu D. A machine-learning approach for computation of fractional flow reserve from coronary computed tomography.J Appl Physiol (1985). 2016; 121:42–52. doi: 10.1152/japplphysiol.00752.2015CrossrefMedlineGoogle Scholar
  • 80. Martin SS, Mastrodicasa D, van Assen M, De Cecco CN, Bayer RR, Tesche C, Varga-Szemes A, Fischer AM, Jacobs BE, Sahbaee Pet al. Value of machine learning-based coronary CT fractional flow reserve applied to triple-rule-out CT angiography in acute chest pain.Radiol Cardiothorac Imaging. 2020; 2:e190137. doi: 10.1148/ryct.2020190137CrossrefMedlineGoogle Scholar
  • 81. Kumamaru KK, Fujimoto S, Otsuka Y, Kawasaki T, Kawaguchi Y, Kato E, Takamura K, Aoshima C, Kamo Y, Kogure Yet al. Diagnostic accuracy of 3D deep-learning-based fully automated estimation of patient-level minimum fractional flow reserve from coronary computed tomography angiography.Eur Heart J Cardiovasc Imaging. 2020; 21:437–445. doi: 10.1093/ehjci/jez160CrossrefMedlineGoogle Scholar
  • 82. Coenen A, Kim YH, Kruk M, Tesche C, De Geer J, Kurata A, Lubbers ML, Daemen J, Itu L, Rapaka Set al. Diagnostic accuracy of a machine-learning approach to coronary computed tomographic angiography-based fractional flow reserve: result from the MACHINE consortium.Circ Cardiovasc Imaging. 2018; 11:e007217. doi: 10.1161/CIRCIMAGING.117.007217LinkGoogle Scholar
  • 83. Baumann S, Renker M, Schoepf UJ, De Cecco CN, Coenen A, De Geer J, Kruk M, Kim YH, Albrecht MH, Duguay TMet al. Gender differences in the diagnostic performance of machine learning coronary CT angiography-derived fractional flow reserve -results from the MACHINE registry.Eur J Radiol. 2019; 119:108657. doi: 10.1016/j.ejrad.2019.108657CrossrefMedlineGoogle Scholar
  • 84. Baumann S, Hirt M, Rott C, Özdemir GH, Tesche C, Becher T, Weiss C, Hetjens S, Akin I, Schoenberg SOet al. Comparison of machine learning computed tomography-based fractional flow reserve and coronary CT angiography-derived plaque characteristics with invasive resting full-cycle ratio.J Clin Med. 2020; 9:E714. doi: 10.3390/jcm9030714CrossrefMedlineGoogle Scholar
  • 85. Tesche C, De Cecco CN, Baumann S, Renker M, McLaurin TW, Duguay TM, Bayer RR, Steinberg DH, Grant KL, Canstein Cet al. Coronary CT angiography-derived fractional flow reserve: machine learning algorithm versus computational fluid dynamics modeling.Radiology. 2018; 288:64–72. doi: 10.1148/radiol.2018171291CrossrefMedlineGoogle Scholar
  • 86. von Knebel Doeberitz PL, De Cecco CN, Schoepf UJ, Duguay TM, Albrecht MH, van Assen M, Bauer MJ, Savage RH, Pannell JT, De Santis Det al. Coronary CT angiography-derived plaque quantification with artificial intelligence CT fractional flow reserve for the identification of lesion-specific ischemia.Eur Radiol. 2019; 29:2378–2387. doi: 10.1007/s00330-018-5834-zCrossrefMedlineGoogle Scholar
  • 87. Tesche C, Otani K, De Cecco CN, Coenen A, De Geer J, Kruk M, Kim YH, Albrecht MH, Baumann S, Renker Met al. Influence of coronary calcium on diagnostic performance of machine learning CT-FFR: results from MACHINE registry.JACC Cardiovasc Imaging. 2020; 13:760–770. doi: 10.1016/j.jcmg.2019.06.027CrossrefMedlineGoogle Scholar
  • 88. Liu X, Mo X, Zhang H, Yang G, Shi C, Hau WK. A 2-year investigation of the impact of the computed tomography-derived fractional flow reserve calculated using a deep learning algorithm on routine decision-making for coronary artery disease management.Eur Radiol. 2021; 31:7039–7046. doi: 10.1007/s00330-021-07771-7CrossrefMedlineGoogle Scholar
  • 89. Tang CX, Guo BJ, Schoepf JU, Bayer RR, Liu CY, Qiao HY, Zhou F, Lu GM, Zhou CS, Zhang LJ. Feasibility and prognostic role of machine learning-based FFRCT in patients with stent implantation.Eur Radiol. 2021; 31:6592–6604. doi: 10.1007/s00330-021-07922-wCrossrefMedlineGoogle Scholar
  • 90. Bai W, Sinclair M, Tarroni G, Oktay O, Rajchl M, Vaillant G, Lee AM, Aung N, Lukaschuk E, Sanghvi MMet al. Automated cardiovascular magnetic resonance image analysis with fully convolutional networks.J Cardiovasc Magn Reson. 2018; 20:65. doi: 10.1186/s12968-018-0471-xCrossrefMedlineGoogle Scholar
  • 91. Tao Q, Yan W, Wang Y, Paiman EHM, Shamonin DP, Garg P, Plein S, Huang L, Xia L, Sramko Met al. Deep learning-based method for fully automatic quantification of left ventricle function from cine MR images: a multivendor, multicenter study.Radiology. 2019; 290:81–88. doi: 10.1148/radiol.2018180513CrossrefMedlineGoogle Scholar
  • 92. Ruijsink B, Puyol-Antón E, Oksuz I, Sinclair M, Bai W, Schnabel JA, Razavi R, King AP. Fully automated, quality-controlled cardiac analysis from CMR: validation and large-scale application to characterize cardiac function.JACC Cardiovasc Imaging. 2020; 13:684–695. doi: 10.1016/j.jcmg.2019.05.030CrossrefMedlineGoogle Scholar
  • 93. Zabihollahy F, White JA, Ukwatta E. Convolutional neural network-based approach for segmentation of left ventricle myocardial scar from 3D late gadolinium enhancement MR images.Med Phys. 2019; 46:1740–1751. doi: 10.1002/mp.13436CrossrefMedlineGoogle Scholar
  • 94. Moccia S, Banali R, Martini C, Muscogiuri G, Pontone G, Pepi M, Caiani EG. Development and testing of a deep learning-based strategy for scar segmentation on CMR-LGE images.MAGMA. 2019; 32:187–195. doi: 10.1007/s10334-018-0718-4CrossrefMedlineGoogle Scholar
  • 95. Zhang N, Yang G, Gao Z, Xu C, Zhang Y, Shi R, Keegan J, Xu L, Zhang H, Fan Z, Firmin D. Deep learning for diagnosis of chronic myocardial infarction on nonenhanced cardiac cine MRI.Radiology. 2019; 291:606–617. doi: 10.1148/radiol.2019182304CrossrefMedlineGoogle Scholar
  • 96. Khurshid S, Friedman S, Pirruccello JP, Di Achille P, Diamant N, Anderson CD, Ellinor PT, Batra P, Ho JE, Philippakis AAet al. Deep learning to predict cardiac magnetic resonance-derived left ventricular mass and hypertrophy from 12-lead ECGs.Circ Cardiovasc Imaging. 2021; 14:e012281. doi: 10.1161/CIRCIMAGING.120.012281LinkGoogle Scholar
  • 97. Kim YC, Kim KR, Choe YH. Automatic myocardial segmentation in dynamic contrast enhanced perfusion MRI using Monte Carlo dropout in an encoder-decoder convolutional neural network.Comput Methods Programs Biomed. 2020; 185:105150. doi: 10.1016/j.cmpb.2019.105150CrossrefMedlineGoogle Scholar
  • 98. Xue H, Davies RH, Brown LAE, Knott KD, Kotecha T, Fontana M, Plein S, Moon JC, Kellman P. Automated inline analysis of myocardial perfusion MRI with deep learning.Radiol Artif Intell. 2020; 2:e200009. doi: 10.1148/ryai.2020200009CrossrefMedlineGoogle Scholar
  • 99. Knott KD, Seraphim A, Augusto JB, Xue H, Chacko L, Aung N, Petersen SE, Cooper JA, Manisty C, Bhuva ANet al. The prognostic significance of quantitative myocardial perfusion: an artificial intelligence-based approach using perfusion mapping.Circulation. 2020; 141:1282–1291. doi: 10.1161/CIRCULATIONAHA.119.044666LinkGoogle Scholar