Automated Quality-Controlled Cardiovascular Magnetic Resonance Pericardial Fat Quantification Using a Convolutional Neural Network in the UK Biobank

DEep LearnIng-based QuaNtification of epicardial adipose tissue predicts MACE in patients undergoing stress CMR

BACKGROUND AND AIMS

This study investigated the additional prognostic value of epicardial adipose tissue (EAT) volume for major adverse cardiovascular events (MACE) in patients undergoing stress cardiac magnetic resonance (CMR) imaging.

METHODS

730 consecutive patients [mean age: 63 ± 10 years; 616 men] who underwent stress CMR for known or suspected coronary artery disease were randomly divided into derivation (n = 365) and validation (n = 365) cohorts. MACE was defined as non-fatal myocardial infarction and cardiac deaths. A deep learning algorithm was developed and trained to quantify EAT volume from CMR. EAT volume was adjusted for height (EAT volume index). A composite CMR-based risk score by Cox analysis of the risk of MACE was created.

RESULTS

In the derivation cohort, 32 patients (8.7 %) developed MACE during a follow-up of 2103 days. Left ventricular ejection fraction (LVEF) < 35 % (HR 4.407 [95 % CI 1.903-10.202]; p<0.001), stress perfusion defect (HR 3.550 [95 % CI 1.765-7.138]; p<0.001), late gadolinium enhancement (LGE) (HR 4.428 [95%CI 1.822-10.759]; p = 0.001) and EAT volume index (HR 1.082 [95 % CI 1.045-1.120]; p<0.001) were independent predictors of MACE. In a multivariate Cox regression analysis, adding EAT volume index to a composite risk score including LVEF, stress perfusion defect and LGE provided additional value in MACE prediction, with a net reclassification improvement of 0.683 (95%CI, 0.336-1.03; p<0.001). The combined evaluation of risk score and EAT volume index showed a higher Harrel C statistic as compared to risk score (0.85 vs. 0.76; p<0.001) and EAT volume index alone (0.85 vs.0.74; p<0.001). These findings were confirmed in the validation cohort.

CONCLUSIONS

In patients with clinically indicated stress CMR, fully automated EAT volume measured by deep learning can provide additional prognostic information on top of standard clinical and imaging parameters.

CT- and MRI-Aided Fluorescence Tomography Reconstructions for Biodistribution Analysis

OBJECTIVES

Optical fluorescence imaging can track the biodistribution of fluorophore-labeled drugs, nanoparticles, and antibodies longitudinally. In hybrid computed tomography-fluorescence tomography (CT-FLT), CT provides the anatomical information to generate scattering and absorption maps supporting a 3-dimensional reconstruction from the raw optical data. However, given the CT's limited soft tissue contrast, fluorescence reconstruction and quantification can be inaccurate and not sufficiently detailed. Magnetic resonance imaging (MRI) can overcome these limitations and extend the options for tissue characterization. Thus, we aimed to establish a hybrid CT-MRI-FLT approach for whole-body imaging and compared it with CT-FLT.

MATERIALS AND METHODS

The MRI-based hybrid imaging approaches were established first by scanning a water and coconut oil-filled phantom, second by quantifying Cy7 concentrations of inserts in dead mice, and finally by analyzing the biodistribution of AF750-labeled immunoglobulins (IgG, IgA) in living SKH1 mice. Magnetic resonance imaging, acquired with a fat-water-separated mDixon sequence, CT, and FLT were co-registered using markers in the mouse holder frame filled with white petrolatum, which was solid, stable, and visible in both modalities.

RESULTS

Computed tomography-MRI fusion was confirmed by comparing the segmentation agreement using Dice scores. Phantom segmentations showed good agreement, after correction for gradient linearity distortion and chemical shift. Organ segmentations in dead and living mice revealed adequate agreement for fusion. Marking the mouse holder frame and the successful CT-MRI fusion enabled MRI-FLT as well as CT-MRI-FLT reconstructions. Fluorescence tomography reconstructions supported by CT, MRI, or CT-MRI were comparable in dead mice with 60 pmol fluorescence inserts at different locations. Although standard CT-FLT reconstruction only considered general values for soft tissue, skin, lung, fat, and bone scattering, MRI's more versatile soft tissue contrast enabled the additional consideration of liver, kidneys, and brain. However, this did not change FLT reconstructions and quantifications significantly, whereas for extending scattering maps, it was important to accurately segment the organs and the entire mouse body. The various FLT reconstructions also provided comparable results for the in vivo biodistribution analyses with fluorescent immunoglobulins. However, MRI additionally enabled the visualization of gallbladder, thyroid, and brain. Furthermore, segmentations of liver, spleen, and kidney were more reliable due to better-defined contours than in CT. Therefore, the improved segmentations enabled better assignment of fluorescence signals and more differentiated conclusions with MRI-FLT.

CONCLUSIONS

Whole-body CT-MRI-FLT was implemented as a novel trimodal imaging approach, which allowed to more accurately assign fluorescence signals, thereby significantly improving pharmacokinetic analyses.

Radiomics of pericardial fat: a new frontier in heart failure discrimination and prediction

OBJECTIVES

To use pericardial adipose tissue (PAT) radiomics phenotyping to differentiate existing and predict future heart failure (HF) cases in the UK Biobank.

METHODS

PAT segmentations were derived from cardiovascular magnetic resonance (CMR) studies using an automated quality-controlled model to define the region-of-interest for radiomics analysis. Prevalent (present at time of imaging) and incident (first occurrence after imaging) HF were ascertained using health record linkage. We created balanced cohorts of non-HF individuals for comparison. PyRadiomics was utilised to extract 104 radiomics features, of which 28 were chosen after excluding highly correlated ones (0.8). These features, plus sex and age, served as predictors in binary classification models trained separately to detect (1) prevalent and (2) incident HF. We tested seven modeling methods using tenfold nested cross-validation and examined feature importance with explainability methods.

RESULTS

We studied 1204 participants in total, 297 participants with prevalent (60 ± 7 years, 21% female) and 305 with incident (61 ± 6 years, 32% female) HF, and an equal number of non-HF comparators. We achieved good discriminative performance for both prevalent (voting classifier; AUC: 0.76; F1 score: 0.70) and incident (light gradient boosting machine: AUC: 0.74; F1 score: 0.68) HF. Our radiomics models showed marginally better performance compared to PAT area alone. Increased PAT size (maximum 2D diameter in a given column or slice) and texture heterogeneity (sum entropy) were important features for prevalent and incident HF classification models.

CONCLUSIONS

The amount and character of PAT discriminate individuals with prevalent HF and predict incidence of future HF.

CLINICAL RELEVANCE STATEMENT

This study presents an innovative application of pericardial adipose tissue (PAT) radiomics phenotyping as a predictive tool for heart failure (HF), a major public health concern. By leveraging advanced machine learning methods, the research uncovers that the quantity and characteristics of PAT can be used to identify existing cases of HF and predict future occurrences. The enhanced performance of these radiomics models over PAT area alone supports the potential for better personalised care through earlier detection and prevention of HF.

KEY POINTS

•PAT radiomics applied to CMR was used for the first time to derive binary machine learning classifiers to develop models for discrimination of prevalence and prediction of incident heart failure. •Models using PAT area provided acceptable discrimination between cases of prevalent or incident heart failure and comparator groups. •An increased PAT volume (increased diameter using shape features) and greater texture heterogeneity captured by radiomics texture features (increased sum entropy) can be used as an additional classifier marker for heart failure.

Genome-Wide Association Study of Pericardial Fat Area in 28 161 UK Biobank Participants

BACKGROUND Pericardial adipose tissue (PAT) is the visceral adipose tissue compartment surrounding the heart. Experimental and observational research has suggested that greater PAT deposition might mediate cardiovascular disease, independent of general or subcutaneous adiposity. We characterize the genetic architecture of adiposity-adjusted PAT and identify causal associations between PAT and adverse cardiac magnetic resonance imaging measures of cardiac structure and function in 28 161 UK Biobank participants. METHODS AND RESULTS The PAT phenotype was extracted from cardiac magnetic resonance images using an automated image analysis tool previously developed and validated in this cohort. A genome-wide association study was performed with PAT area set as the phenotype, adjusting for age, sex, and other measures of obesity. Functional mapping and Bayesian colocalization were used to understand the biologic role of identified variants. Mendelian randomization analysis was used to examine potential causal links between genetically determined PAT and cardiac magnetic resonance-derived measures of left ventricular structure and function. We discovered 12 genome-wide significant variants, with 2 independent sentinel variants (rs6428792, P=4.20×10-9 and rs11992444, P=1.30×10-12) at 2 distinct genomic loci, that were mapped to 3 potentially causal genes: T-box transcription factor 15 (TBX15), tryptophanyl tRNA synthetase 2, mitochondrial (WARS2) and early B-cell factor-2 (EBF2) through functional annotation. Bayesian colocalization additionally suggested a role of RP4-712E4.1. Genetically predicted differences in adiposity-adjusted PAT were causally associated with adverse left ventricular remodeling. CONCLUSIONS This study provides insights into the genetic architecture determining differential PAT deposition, identifies causal links with left structural and functional parameters, and provides novel data about the pathophysiological importance of adiposity distribution.

Multilevel comparison of deep learning models for function quantification in cardiovascular magnetic resonance: On the redundancy of architectural variations

Background

Cardiac function quantification in cardiovascular magnetic resonance requires precise contouring of the heart chambers. This time-consuming task is increasingly being addressed by a plethora of ever more complex deep learning methods. However, only a small fraction of these have made their way from academia into clinical practice. In the quality assessment and control of medical artificial intelligence, the opaque reasoning and associated distinctive errors of neural networks meet an extraordinarily low tolerance for failure.

Aim

The aim of this study is a multilevel analysis and comparison of the performance of three popular convolutional neural network (CNN) models for cardiac function quantification.

Methods

U-Net, FCN, and MultiResUNet were trained for the segmentation of the left and right ventricles on short-axis cine images of 119 patients from clinical routine. The training pipeline and hyperparameters were kept constant to isolate the influence of network architecture. CNN performance was evaluated against expert segmentations for 29 test cases on contour level and in terms of quantitative clinical parameters. Multilevel analysis included breakdown of results by slice position, as well as visualization of segmentation deviations and linkage of volume differences to segmentation metrics via correlation plots for qualitative analysis.

Results

All models showed strong correlation to the expert with respect to quantitative clinical parameters (r_{z '} = 0.978, 0.977, 0.978 for U-Net, FCN, MultiResUNet respectively). The MultiResUNet significantly underestimated ventricular volumes and left ventricular myocardial mass. Segmentation difficulties and failures clustered in basal and apical slices for all CNNs, with the largest volume differences in the basal slices (mean absolute error per slice: 4.2 ± 4.5 ml for basal, 0.9 ± 1.3 ml for midventricular, 0.9 ± 0.9 ml for apical slices). Results for the right ventricle had higher variance and more outliers compared to the left ventricle. Intraclass correlation for clinical parameters was excellent (≥0.91) among the CNNs.

Conclusion

Modifications to CNN architecture were not critical to the quality of error for our dataset. Despite good overall agreement with the expert, errors accumulated in basal and apical slices for all models.

Clinician's guide to trustworthy and responsible artificial intelligence in cardiovascular imaging

A growing number of artificial intelligence (AI)-based systems are being proposed and developed in cardiology, driven by the increasing need to deal with the vast amount of clinical and imaging data with the ultimate aim of advancing patient care, diagnosis and prognostication. However, there is a critical gap between the development and clinical deployment of AI tools. A key consideration for implementing AI tools into real-life clinical practice is their "trustworthiness" by end-users. Namely, we must ensure that AI systems can be trusted and adopted by all parties involved, including clinicians and patients. Here we provide a summary of the concepts involved in developing a "trustworthy AI system." We describe the main risks of AI applications and potential mitigation techniques for the wider application of these promising techniques in the context of cardiovascular imaging. Finally, we show why trustworthy AI concepts are important governing forces of AI development.

Pericardial adiposity is independently linked to adverse cardiovascular phenotypes: a CMR study of 42 598 UK Biobank participants

AIMS

We evaluated independent associations of cardiovascular magnetic resonance (CMR)-measured pericardial adipose tissue (PAT) with cardiovascular structure and function and considered underlying mechanism in 42 598 UK Biobank participants.

METHODS AND RESULTS

We extracted PAT and selected CMR metrics using automated pipelines. We estimated associations of PAT with each CMR metric using linear regression adjusting for age, sex, ethnicity, deprivation, smoking, exercise, processed food intake, body mass index, diabetes, hypertension, height cholesterol, waist-to-hip ratio, impedance fat measures, and magnetic resonance imaging abdominal visceral adiposity measures. Higher PAT was independently associated with unhealthy left ventricular (LV) structure (greater wall thickness, higher LV mass, more concentric pattern of LV hypertrophy), poorer LV function (lower LV global function index, lower LV stroke volume), lower left atrial ejection fraction, and lower aortic distensibility. We used multiple mediation analysis to examine the potential mediating effect of cardiometabolic diseases and blood biomarkers (lipid profile, glycaemic control, inflammation) in the PAT-CMR relationships. Higher PAT was associated with cardiometabolic disease (hypertension, diabetes, high cholesterol), adverse serum lipids, poorer glycaemic control, and greater systemic inflammation. We identified potential mediation pathways via hypertension, adverse lipids, and inflammation markers, which overall only partially explained the PAT-CMR relationships.

CONCLUSION

We demonstrate association of PAT with unhealthy cardiovascular structure and function, independent of baseline comorbidities, vascular risk factors, inflammatory markers, and multiple non-invasive and imaging measures of obesity. Our findings support an independent role of PAT in adversely impacting cardiovascular health and highlight CMR-measured PAT as a potential novel imaging biomarker of cardiovascular risk.

A Survey on the Role of Artificial Intelligence in Biobanking Studies: A Systematic Review

Introduction: In biobanks, participants' biological samples are stored for future research. The application of artificial intelligence (AI) involves the analysis of data and the prediction of any pathological outcomes. In AI, models are used to diagnose diseases as well as classify and predict disease risks. Our research analyzed AI's role in the development of biobanks in the healthcare industry, systematically. Methods: The literature search was conducted using three digital reference databases, namely PubMed, CINAHL, and WoS. Guidelines for preferred reporting elements for systematic reviews and meta-analyses (PRISMA)-2020 in conducting the systematic review were followed. The search terms included "biobanks", "AI", "machine learning", and "deep learning", as well as combinations such as "biobanks with AI", "deep learning in the biobanking field", and "recent advances in biobanking". Only English-language papers were included in the study, and to assess the quality of selected works, the Newcastle-Ottawa scale (NOS) was used. The good quality range (NOS ≥ 7) is only considered for further review. Results: A literature analysis of the above entries resulted in 239 studies. Based on their relevance to the study's goal, research characteristics, and NOS criteria, we included 18 articles for reviewing. In the last decade, biobanks and artificial intelligence have had a relatively large impact on the medical system. Interestingly, UK biobanks account for the highest percentage of high-quality works, followed by Qatar, South Korea, Singapore, Japan, and Denmark. Conclusions: Translational bioinformatics probably represent a future leader in precision medicine. AI and machine learning applications to biobanking research may contribute to the development of biobanks for the utility of health services and citizens.

Cardiac MR: From Theory to Practice

Cardiovascular disease (CVD) is the leading single cause of morbidity and mortality, causing over 17. 9 million deaths worldwide per year with associated costs of over $800 billion. Improving prevention, diagnosis, and treatment of CVD is therefore a global priority. Cardiovascular magnetic resonance (CMR) has emerged as a clinically important technique for the assessment of cardiovascular anatomy, function, perfusion, and viability. However, diversity and complexity of imaging, reconstruction and analysis methods pose some limitations to the widespread use of CMR. Especially in view of recent developments in the field of machine learning that provide novel solutions to address existing problems, it is necessary to bridge the gap between the clinical and scientific communities. This review covers five essential aspects of CMR to provide a comprehensive overview ranging from CVDs to CMR pulse sequence design, acquisition protocols, motion handling, image reconstruction and quantitative analysis of the obtained data. (1) The basic MR physics of CMR is introduced. Basic pulse sequence building blocks that are commonly used in CMR imaging are presented. Sequences containing these building blocks are formed for parametric mapping and functional imaging techniques. Commonly perceived artifacts and potential countermeasures are discussed for these methods. (2) CMR methods for identifying CVDs are illustrated. Basic anatomy and functional processes are described to understand the cardiac pathologies and how they can be captured by CMR imaging. (3) The planning and conduct of a complete CMR exam which is targeted for the respective pathology is shown. Building blocks are illustrated to create an efficient and patient-centered workflow. Further strategies to cope with challenging patients are discussed. (4) Imaging acceleration and reconstruction techniques are presented that enable acquisition of spatial, temporal, and parametric dynamics of the cardiac cycle. The handling of respiratory and cardiac motion strategies as well as their integration into the reconstruction processes is showcased. (5) Recent advances on deep learning-based reconstructions for this purpose are summarized. Furthermore, an overview of novel deep learning image segmentation and analysis methods is provided with a focus on automatic, fast and reliable extraction of biomarkers and parameters of clinical relevance.

Deep-Learning Segmentation of Epicardial Adipose Tissue Using Four-Chamber Cardiac Magnetic Resonance Imaging

In magnetic resonance imaging (MRI), epicardial adipose tissue (EAT) overload remains often overlooked due to tedious manual contouring in images. Automated four-chamber EAT area quantification was proposed, leveraging deep-learning segmentation using multi-frame fully convolutional networks (FCN). The investigation involved 100 subjects-comprising healthy, obese, and diabetic patients-who underwent 3T cardiac cine MRI, optimized U-Net and FCN (noted FCNB) were trained on three consecutive cine frames for segmentation of central frame using dice loss. Networks were trained using 4-fold cross-validation (n = 80) and evaluated on an independent dataset (n = 20). Segmentation performances were compared to inter-intra observer bias with dice (DSC) and relative surface error (RSE). Both systole and diastole four-chamber area were correlated with total EAT volume (r = 0.77 and 0.74 respectively). Networks' performances were equivalent to inter-observers' bias (EAT: DSCInter = 0.76, DSCU-Net = 0.77, DSCFCNB = 0.76). U-net outperformed (p < 0.0001) FCNB on all metrics. Eventually, proposed multi-frame U-Net provided automated EAT area quantification with a 14.2% precision for the clinically relevant upper three quarters of EAT area range, scaling patients' risk of EAT overload with 70% accuracy. Exploiting multi-frame U-Net in standard cine provided automated EAT quantification over a wide range of EAT quantities. The method is made available to the community through a FSLeyes plugin.

Epicardial Fat Expansion in Diabetic and Obese Patients With Heart Failure and Preserved Ejection Fraction-A Specific HFpEF Phenotype

Heart failure with preserved ejection fraction (HFpEF) is a heterogeneous syndrome with diverse etiologies and pathophysiological factors. Obesity and type 2 diabetes mellitus (T2DM), conditions that coexist frequently, induce a cluster of metabolic and non-metabolic signaling derangements which are in favor to induce inflammation, fibrosis, myocyte stiffness, all hallmarks of HFpEF. In contrast to other HFpEF risk factors, obesity and T2DM are often associated with the generation of enlarged epicardial adipose tissue (EAT). EAT acts as an endocrine tissue that may exacerbate myocardial inflammation and fibrosis via various paracrine and vasocrine signals. In addition, an abnormally large EAT poses mechanical stress on the heart via pericardial restrain. HFpEF patients with enlarged EAT may belong to a unique phenotype that can benefit from specific EAT-targeted interventions, including life-style modifications and pharmacologically via statins and fat modifying anti-diabetics drugs; like metformin, sodium-glucose cotransporter 2 inhibitors, or glucagon-like peptide-1 receptor agonists, respectively.