Artificial Intelligence in Cardiac Imaging With Statistical Atlases of Cardiac Anatomy

CHeart: A Conditional Spatio-Temporal Generative Model for Cardiac Anatomy

Two key questions in cardiac image analysis are to assess the anatomy and motion of the heart from images; and to understand how they are associated with non-imaging clinical factors such as gender, age and diseases. While the first question can often be addressed by image segmentation and motion tracking algorithms, our capability to model and answer the second question is still limited. In this work, we propose a novel conditional generative model to describe the 4D spatio-temporal anatomy of the heart and its interaction with non-imaging clinical factors. The clinical factors are integrated as the conditions of the generative modelling, which allows us to investigate how these factors influence the cardiac anatomy. We evaluate the model performance in mainly two tasks, anatomical sequence completion and sequence generation. The model achieves high performance in anatomical sequence completion, comparable to or outperforming other state-of-the-art generative models. In terms of sequence generation, given clinical conditions, the model can generate realistic synthetic 4D sequential anatomies that share similar distributions with the real data. The code and the trained generative model are available at https://github.com/MengyunQ/CHeart.

Machine learning used for simulation of MitraClip intervention: A proof-of-concept study

Introduction: Severe mitral regurgitation (MR) is a mitral valve disease that can lead to lifethreatening complications. MitraClip (MC) therapy is a percutaneous solution for patients who cannot tolerate surgical solutions. In MC therapy, a clip is implanted in the heart to reduce MR. To achieve optimal MC therapy, the cardiologist needs to foresee the outcomes of different scenarios for MC implantation, including the location of the MC. Although finite element (FE) modeling can simulate the outcomes of different MC scenarios, it is not suitable for clinical usage because it requires several hours to complete.

Methods: In this paper, we used machine learning (ML) to predict the outcomes of MC therapy in less than 1 s. Two ML algorithms were used: XGBoost, which is a decision tree model, and a feed-forward deep learning (DL) model. The MC location, the geometrical attributes of the models and baseline stress and MR were the features of the ML models, and the predictions were performed for MR and maximum von Mises stress in the leaflets. The parameters of the ML models were determined to achieve the minimum errors obtained by applying the ML models on the validation set.

Results: The results for the test set (not used during training) showed relative agreement between ML predictions and ground truth FE predictions. The accuracy of the XGBoost models were better than DL models. Mean absolute percentage error (MAPE) for the XGBoost predictions were 0.115 and 0.231, and the MAPE for DL predictions were 0.154 and 0.310, for MR and stress, respectively.

Discussion: The ML models reduced the FE runtime from 6 hours (on average) to less than 1 s. The accuracy of ML models can be increased by increasing the dataset size. The results of this study have important implications for improving the outcomes of MC therapy by providing information about the outcomes of MC implantation in real-time.

Predicting one-year left ventricular mass index regression following transcatheter aortic valve replacement in patients with severe aortic stenosis: A new era is coming

Aortic stenosis (AS) is the most common valvular heart disease in the western world, particularly worrisome with an ever-aging population wherein postoperative outcome for aortic valve replacement is strongly related to the timing of surgery in the natural course of disease. Yet, guidelines for therapy planning overlook insightful, quantified measures from medical imaging to educate clinical decisions. Herein, we leverage statistical shape analysis (SSA) techniques combined with customized machine learning methods to extract latent information from segmented left ventricle (LV) shapes. This enabled us to predict left ventricular mass index (LVMI) regression a year after transcatheter aortic valve replacement (TAVR). LVMI regression is an expected phenomena in patients undergone aortic valve replacement reported to be tightly correlated with survival one and five year after the intervention. In brief, LV geometries were extracted from medical images of a cohort of AS patients using deep learning tools, and then analyzed to create a set of statistical shape models (SSMs). Then, the supervised shape features were extracted to feed a support vector regression (SVR) model to predict the LVMI regression. The average accuracy of the predictions was validated against clinical measurements calculating root mean square error and R 2 score which yielded the satisfactory values of 0.28 and 0.67, respectively, on test data. Our work reveals the promising capability of advanced mathematical and bioinformatics approaches such as SSA and machine learning to improve medical output prediction and treatment planning.

Interpretable cardiac anatomy modeling using variational mesh autoencoders

Cardiac anatomy and function vary considerably across the human population with important implications for clinical diagnosis and treatment planning. Consequently, many computer-based approaches have been developed to capture this variability for a wide range of applications, including explainable cardiac disease detection and prediction, dimensionality reduction, cardiac shape analysis, and the generation of virtual heart populations. In this work, we propose a variational mesh autoencoder (mesh VAE) as a novel geometric deep learning approach to model such population-wide variations in cardiac shapes. It embeds multi-scale graph convolutions and mesh pooling layers in a hierarchical VAE framework to enable direct processing of surface mesh representations of the cardiac anatomy in an efficient manner. The proposed mesh VAE achieves low reconstruction errors on a dataset of 3D cardiac meshes from over 1,000 patients with acute myocardial infarction, with mean surface distances between input and reconstructed meshes below the underlying image resolution. We also find that it outperforms a voxelgrid-based deep learning benchmark in terms of both mean surface distance and Hausdorff distance while requiring considerably less memory. Furthermore, we explore the quality and interpretability of the mesh VAE's latent space and showcase its ability to improve the prediction of major adverse cardiac events over a clinical benchmark. Finally, we investigate the method's ability to generate realistic virtual populations of cardiac anatomies and find good alignment between the synthesized and gold standard mesh populations in terms of multiple clinical metrics.

Artificial intelligence-based opportunistic screening for the detection of arterial hypertension through ECG signals

OBJECTIVES

Hypertension is a major risk factor for cardiovascular disease (CVD), which often escapes the diagnosis or should be confirmed by several office visits. The ECG is one of the most widely used diagnostic tools and could be of paramount importance in patients' initial evaluation.

METHODS

We used machine learning techniques based on clinical parameters and features derived from the ECG, to detect hypertension in a population without CVD. We enrolled 1091 individuals who were classified as hypertensive or normotensive, and trained a Random Forest model, to detect the existence of hypertension. We then calculated the values for the Shapley additive explanations (SHAP), a sophisticated feature importance analysis, to interpret each feature's role in the Random Forest's results.

RESULTS

Our Random Forest model was able to distinguish hypertensive from normotensive patients with accuracy 84.2%, specificity 78.0%, sensitivity 84.0% and area under the receiver-operating curve 0.89, using a decision threshold of 0.6. Age, BMI, BMI-adjusted Cornell criteria (BMI multiplied by RaVL+SV 3 ), R wave amplitude in aVL and BMI-modified Sokolow-Lyon voltage (BMI divided by SV 1 +RV 5 ), were the most important anthropometric and ECG-derived features in terms of the success of our model.

CONCLUSION

Our machine learning algorithm is effective in the detection of hypertension in patients using ECG-derived and basic anthropometric criteria. Our findings open new horizon in the detection of many undiagnosed hypertensive individuals who have an increased CVD risk.

Le Cœur en Sabot: shape associations with adverse events in repaired tetralogy of Fallot

BACKGROUND

Maladaptive remodelling mechanisms occur in patients with repaired tetralogy of Fallot (rToF) resulting in a cycle of metabolic and structural changes. Biventricular shape analysis may indicate mechanisms associated with adverse events independent of pulmonary regurgitant volume index (PRVI). We aimed to determine novel remodelling patterns associated with adverse events in patients with rToF using shape and function analysis.

METHODS

Biventricular shape and function were studied in 192 patients with rToF (median time from TOF repair to baseline evaluation 13.5 years). Linear discriminant analysis (LDA) and principal component analysis (PCA) were used to identify shape differences between patients with and without adverse events. Adverse events included death, arrhythmias, and cardiac arrest with median follow-up of 10 years.

RESULTS

LDA and PCA showed that shape characteristics pertaining to adverse events included a more circular left ventricle (LV) (decreased eccentricity), dilated (increased sphericity) LV base, increased right ventricular (RV) apical sphericity, and decreased RV basal sphericity. Multivariate LDA showed that the optimal discriminative model included only RV apical ejection fraction and one PCA mode associated with a more circular and dilated LV base (AUC = 0.77). PRVI did not add value, and shape changes associated with increased PRVI were not predictive of adverse outcomes.

CONCLUSION

Pathological remodelling patterns in patients with rToF are significantly associated with adverse events, independent of PRVI. Mechanisms related to incident events include LV basal dilation with a reduced RV apical ejection fraction.

CT-Based Analysis of Left Ventricular Hemodynamics Using Statistical Shape Modeling and Computational Fluid Dynamics

Background

Cardiac computed tomography (CCT) based computational fluid dynamics (CFD) allows to assess intracardiac flow features, which are hypothesized as an early predictor for heart diseases and may support treatment decisions. However, the understanding of intracardiac flow is challenging due to high variability in heart shapes and contractility. Using statistical shape modeling (SSM) in combination with CFD facilitates an intracardiac flow analysis. The aim of this study is to prove the usability of a new approach to describe various cohorts.

Materials and Methods

CCT data of 125 patients (mean age: 60.6 ± 10.0 years, 16.8% woman) were used to generate SSMs representing aneurysmatic and non-aneurysmatic left ventricles (LVs). Using SSMs, seven group-averaged LV shapes and contraction fields were generated: four representing patients with and without aneurysms and with mild or severe mitral regurgitation (MR), and three distinguishing aneurysmatic patients with true, intermediate aneurysms, and globally hypokinetic LVs. End-diastolic LV volumes of the groups varied between 258 and 347 ml, whereas ejection fractions varied between 21 and 26%. MR degrees varied from 1.0 to 2.5. Prescribed motion CFD was used to simulate intracardiac flow, which was analyzed regarding large-scale flow features, kinetic energy, washout, and pressure gradients.

Results

SSMs of aneurysmatic and non-aneurysmatic LVs were generated. Differences in shapes and contractility were found in the first three shape modes. Ninety percent of the cumulative shape variance is described with approximately 30 modes. A comparison of hemodynamics between all groups found shape-, contractility- and MR-dependent differences. Disturbed blood washout in the apex region was found in the aneurysmatic cases. With increasing MR, the diastolic jet becomes less coherent, whereas energy dissipation increases by decreasing kinetic energy. The poorest blood washout was found for the globally hypokinetic group, whereas the weakest blood washout in the apex region was found for the true aneurysm group.

Conclusion

The proposed CCT-based analysis of hemodynamics combining CFD with SSM seems promising to facilitate the analysis of intracardiac flow, thus increasing the value of CCT for diagnostic and treatment decisions. With further enhancement of the computational approach, the methodology has the potential to be embedded in clinical routine workflows and support clinicians.

Multi-Domain Variational Autoencoders for Combined Modeling of MRI-Based Biventricular Anatomy and ECG-Based Cardiac Electrophysiology

Human cardiac function is characterized by a complex interplay of mechanical deformation and electrophysiological conduction. Similar to the underlying cardiac anatomy, these interconnected physiological patterns vary considerably across the human population with important implications for the effectiveness of clinical decision-making and the accuracy of computerized heart models. While many previous works have investigated this variability separately for either cardiac anatomy or physiology, this work aims to combine both aspects in a single data-driven approach and capture their intricate interdependencies in a multi-domain setting. To this end, we propose a novel multi-domain Variational Autoencoder (VAE) network to capture combined Electrocardiogram (ECG) and Magnetic Resonance Imaging (MRI)-based 3D anatomy information in a single model. Each VAE branch is specifically designed to address the particular challenges of the respective input domain, enabling efficient encoding, reconstruction, and synthesis of multi-domain cardiac signals. Our method achieves high reconstruction accuracy on a United Kingdom Biobank dataset, with Chamfer Distances between reconstructed and input anatomies below the underlying image resolution and ECG reconstructions outperforming multiple single-domain benchmarks by a considerable margin. The proposed VAE is capable of generating realistic virtual populations of arbitrary size with good alignment in clinical metrics between the synthesized and gold standard anatomies and Maximum Mean Discrepancy (MMD) scores of generated ECGs below those of comparable single-domain approaches. Furthermore, we observe the latent space of our VAE to be highly interpretable with separate components encoding different aspects of anatomical and ECG variability. Finally, we demonstrate that the combined anatomy and ECG representation improves the performance in a cardiac disease classification task by 3.9% in terms of Area Under the Receiver Operating Characteristic (AUROC) curve over the best corresponding single-domain modeling approach.

Feasibility of a longitudinal statistical atlas model to study aortic growth in congenital heart disease

Studying anatomical shape progression over time is of utmost importance to refine our understanding of clinically relevant processes. These include vascular remodeling, such as aortic dilation, which is particularly important in some congenital heart defects (CHD). A novel methodological framework for three-dimensional shape analysis has been applied for the first time in a CHD scenario, i.e., bicuspid aortic valve (BAV) disease, the most common CHD. Three-dimensional aortic shapes (n = 94) reconstructed from cardiovascular magnetic resonance imaging (MRI) data as surface meshes represented the input for a longitudinal atlas model, using multiple scans over time (n = 2-4 per patient). This model relies on diffeomorphism transformations in the absence of point-to-point correspondence, and on the right combination of initialization, estimation and registration parameters. We computed the shape trajectory of an average disease progression in our cohort, as well as time-dependent parameters, geometric variations and the average shape of the population. Results cover a spatiotemporal spectrum of visual and numerical information that can be further used to run clinical associations. This proof-of-concept study demonstrates the feasibility of applying advanced statistical shape models to track disease progression and stratify patients with CHD.

A deep convolutional neural network for estimating hemodynamic response function with reduction of motion artifacts in fNIRS

OBJECTIVE

Functional near-infrared spectroscopy (fNIRS) is a neuroimaging technique for monitoring hemoglobin concentration changes in a non-invasive manner. However, subject movements are often significant sources of artifacts. While several methods have been developed for suppressing this confounding noise, the conventional techniques have limitations on optimal selections of model parameters across participants or brain regions. To address this shortcoming, we aim to propose a method based on a deep convolutional neural network (CNN).

APPROACH

The U-net is employed as a CNN architecture. Specifically, large-scale training and testing data are generated by combining variants of hemodynamic response function (HRF) with experimental measurements of motion noises. The neural network is then trained to reconstruct hemodynamic response coupled to neuronal activity with a reduction of motion artifacts.

MAIN RESULTS

Using extensive analysis, we show that the proposed method estimates the task-related HRF more accurately than the existing methods of wavelet decomposition and autoregressive models. Specifically, the mean squared error and variance of HRF estimates, based on the CNN, are the smallest among all methods considered in this study. These results are more prominent when the semi-simulated data contains variants of shapes and amplitudes of HRF.

SIGNIFICANCE

The proposed CNN method allows for accurately estimating amplitude and shape of HRF with significant reduction of motion artifacts. This method may have a great potential for monitoring HRF changes in real-life settings that involve excessive motion artifacts.

Characterizing interactions between cardiac shape and deformation by non-linear manifold learning

In clinical routine, high-dimensional descriptors of the cardiac function such as shape and deformation are reduced to scalars (e.g. volumes or ejection fraction), which limit the characterization of complex diseases. Besides, these descriptors undergo interactions depending on disease, which may bias their computational analysis. In this paper, we aim at characterizing such interactions by unsupervised manifold learning. We propose to use a sparsified version of Multiple Manifold Learning to align the latent spaces encoding each descriptor and weighting the strength of the alignment depending on each pair of samples. While this framework was up to now only applied to link different datasets from the same manifold, we demonstrate its relevance to characterize the interactions between different but partially related descriptors of the cardiac function (shape and deformation). We benchmark our approach against linear and non-linear embedding strategies, among which the fusion of manifolds by Multiple Kernel Learning, the independent embedding of each descriptor by Diffusion Maps, and a strict alignment based on pairwise correspondences. We first evaluated the methods on a synthetic dataset from a 0D cardiac model where the interactions between descriptors are fully controlled. Then, we transfered them to a population of right ventricular meshes from 310 subjects (100 healthy and 210 patients with right ventricular disease) obtained from 3D echocardiography, where the link between shape and deformation is key for disease understanding. Our experiments underline the relevance of jointly considering shape and deformation descriptors, and that manifold alignment is preferable over fusion for our application. They also confirm at a finer scale the characteristic traits of the right ventricular diseases in our population.

A probabilistic deep motion model for unsupervised cardiac shape anomaly assessment

Automatic shape anomaly detection in large-scale imaging data can be useful for screening suboptimal segmentations and pathologies altering the cardiac morphology without intensive manual labour. We propose a deep probabilistic model for local anomaly detection in sequences of heart shapes, modelled as point sets, in a cardiac cycle. A deep recurrent encoder-decoder network captures the spatio-temporal dependencies to predict the next shape in the cycle and thus derive the outlier points that are attributed to excessive deviations from the network prediction. A predictive mixture distribution models the inlier and outlier classes via Gaussian and uniform distributions, respectively. A Gibbs sampling Expectation-Maximisation (EM) algorithm computes soft anomaly scores of the points via the posterior probabilities of each class in the E-step and estimates the parameters of the network and the predictive distribution in the M-step. We demonstrate the versatility of the method using two shape datasets derived from: (i) one million biventricular CMR images from 20,000 participants in the UK Biobank (UKB), and (ii) routine diagnostic imaging from Multi-Centre, Multi-Vendor, and Multi-Disease Cardiac Image (M&Ms). Experiments show that the detected shape anomalies in the UKB dataset are mostly associated with poor segmentation quality, and the predicted shape sequences show significant improvement over the input sequences. Furthermore, evaluations on U-Net based shapes from the M&Ms dataset reveals that the anomalies are attributable to the underlying pathologies that affect the ventricles. The proposed model can therefore be used as an effective mechanism to sift shape anomalies in large-scale cardiac imaging pipelines for further analysis.

Atlas-based methods for efficient characterization of patient-specific ventricular activation patterns

AIMS

Ventricular activation patterns can aid clinical decision-making directly by providing spatial information on cardiac electrical activation or indirectly through derived clinical indices. The aim of this work was to derive an atlas of the major modes of variation of ventricular activation from model-predicted 3D bi-ventricular activation time distributions and to relate these modes to corresponding vectorcardiograms (VCGs). We investigated how the resulting dimensionality reduction can improve and accelerate the estimation of activation patterns from surface electrogram measurements.

METHODS AND RESULTS

Atlases of activation time (AT) and VCGs were derived using principal component analysis on a dataset of simulated electrophysiology simulations computed on eight patient-specific bi-ventricular geometries. The atlases provided significant dimensionality reduction, and the modes of variation in the two atlases described similar features. Utility of the atlases was assessed by resolving clinical waveforms against them and the VCG atlas was able to accurately reconstruct the patient VCGs with fewer than 10 modes. A sensitivity analysis between the two atlases was performed by calculating a compact Jacobian. Finally, VCGs generated by varying AT atlas modes were compared with clinical VCGs to estimate patient-specific activation maps, and the resulting errors between the clinical and atlas-based VCGs were less than those from more computationally expensive method.

CONCLUSION

Atlases of activation and VCGs represent a new method of identifying and relating the features of these high-dimensional signals that capture the major sources of variation between patients and may aid in identifying novel clinical indices of arrhythmia risk or therapeutic outcome.

Multiphysics computational modelling of the cardiac ventricles

Development of cardiac multiphysics models has progressed significantly over the decades and simulations combining multiple physics interactions have become increasingly common. In this review, we summarise the progress in this field focusing on various approaches of integrating ventricular structures. electrophysiological properties, myocardial mechanics, as well as incorporating blood hemodynamics and the circulatory system. Common coupling approaches are discussed and compared, including the advantages and shortcomings of each. Currently used strategies for patient-specific implementations are highlighted and potential future improvements considered.

Detection of abnormal left ventricular geometry in patients without cardiovascular disease through machine learning: An ECG-based approach

Cardiac remodeling is recognized as an important aspect of cardiovascular disease (CVD) progression. Machine learning (ML) techniques were applied to basic clinical parameters and electrocardiographic features, in order to detect abnormal left ventricular geometry (LVG) even before the onset of left ventricular hypertrophy (LVH), in a population without established CVD. The authors enrolled 528 patients with and without essential hypertension, but no other indications of CVD. All patients underwent a full echocardiographic evaluation and were classified into 3 groups; normal geometry (NG), concentric remodeling without LVH (CR), and LVH. Abnormal LVG was identified as increased relative wall thickness (RWT) and/or left ventricular mass index (LVMi). The authors trained supervised ML models to classify patients with abnormal LVG and calculated SHAP values to perform feature importance and interaction analysis. Hypertension, age, body mass index over the Sokolow‐Lyon voltage, QRS‐T angle, and QTc duration were some of the most important features. Our model was able to distinguish NG from CR+LVH combined, with 87% accuracy on an unseen test set, 75% specificity, 97% sensitivity, and area under the receiver operating curve (AUC/ROC) equal to 0.91. The authors also trained our model to classify NG and CR (NG + CR) against those with LVH, with 89% test set accuracy, 93% specificity, 67% sensitivity, and an AUC/ROC value of 0.89, for a 0.4 decision threshold. Our ML algorithm effectively detects abnormal LVG even at early stages. Innovative solutions are needed to improve risk stratification of patients without established CVD, and ML may enable progress in this direction.