Structured Abstract Objective To evaluate accuracy and reproducibility of 2D echocardiography (2DE) left ventricular (LV) volumes and ejection fraction (LVEF) estimates by Deep Learning (DL) vs. manual contouring and against CMR. Background 2DE LV manual segmentation for LV volumes and LVEF calculation is time consuming and operator dependent. Methods A DL-based convolutional network (DL1) was trained on 2DE data from centre A, then evaluated on 171 subjects with a wide range of cardiac conditions (49 healthy) – 31 subjects from centre A (18%) and 140 subjects from centre B (82%) – who underwent 2DE and CMR on the same day. Two senior (A 1 and B 1 ) and one junior (A 2 ) cardiologists manually contoured 2DE end-diastolic (ED) and end-systolic (ES) endocardial borders in the cycle and frames of their choice. Selected frames were automatically segmented by DL1 and two DL algorithms from the literature (DL2 and DL3), applied without adaptation to verify their generalizability to unseen data. Interobserver variability of DL was compared to manual contouring. All ESV, EDV and EF values were compared to CMR as reference. Results 50% of 2DE images were of good quality. Interobserver agreement was better by DL1 and DL2 than by manual contouring for EF (Lin’s concordance = 0.9 and 0.91 vs. 0.84), EDV (0.98 and 0.99 vs. 0.82), and ESV (0.99 and 0.99 vs. 0.89). LVEF bias was similar or reduced using DL1 (-0.1) vs. manual contouring (3.0), and worse for DL2 and DL3. Agreement between 2DE and CMR LVEF was similar or higher for DL1 vs. manual contouring (Cohen’s kappa = 0.65 vs. 0.61) and degraded for DL2 and DL3 (0.48 and 0.29). Conclusion DL contouring yielded accurate EF measurements and generalized well to unseen data, while reducing interobserver variability. This suggests that DL contouring may improve accuracy and reproducibility of 2DE LVEF in routine practice.

We introduce a framework for the statistical characterization of heart remodeling from both shape and dynamics of the left ventricle. Shape was characterized by thickness and radius maps, unfolded in a two-dimensional dense Bull’s eye. Motion was represented as a mixture of affine transformations in an anatomical space of coordinates. Using this representation, a population can be projected (after defining spatiotemporal correspondences) to an atlas space built for a given reference population - here, healthy subjects using a classic PCA approach - yielding a joint model of healthy shape and motion statistics. The reconstruction error on shape and motion can then be exploited to quantify remodeling abnormalities. We demonstrate these concepts on 48 healthy subjects and 62 patients with infarct (29 with one year follow-up) imaged with 3D echocardiography, analyzing a total of 139 sequences.

BACKGROUND:In prior work, the authors demonstrated that two-dimensional speckle-tracking (2DST) correlated well but systematically overestimated global longitudinal strain (LS) and circumferential strain (CS) compared with two-dimensional cardiac magnetic resonance tagging (2DTagg) and had poor agreement on a segmental basis. Because three-dimensional speckle-tracking (3DST) has recently emerged as a new, more comprehensive evaluation of myocardial deformation, this study was undertaken to evaluate whether it would compare more favorably with 2DTagg than 2DST. METHODS:In a prospective two-center trial, 119 subjects (29 healthy volunteers, 63 patients with left ventricular dysfunction, and 27 patients with left ventricular hypertrophy) underwent 2DST, 3DST, and 2DTagg. Global, regional (basal, mid, and apical), and segmental (18 and 16 segments per patient) LS and CS by 2DST and 3DST were compared with 2DTagg using intraclass correlation coefficients (ICCs) and Bland-Altman analysis. Test-retest reproducibility of 3DST and 2DST was compared in 48 other patients. RESULTS:Both global LS and CS by 3DST agreed better with 2DTagg (ICC = 0.89 and ICC = 0.83, P < .001 for both; bias = 0.5 ± 2.3% and 0.2 ± 3%) than 2DST (ICC = 0.65 and ICC = 0.55, P < .001 for both; bias = -5.5 ± 2.5% and -7 ± 5.3%). Unlike 2DST, 3DST did not overestimate deformation at the regional and particularly the apical levels and at the segmental level had lower bias (LS, 0.8 ± 2.8% vs -5.3 ± 2.4%; CS, -0.01 ± 2.8% vs -7 ± 2.8%, respectively) but similar agreement with 2DST (LS: ICC = 0.58 ± 0.16 vs 0.56 ± 0.12; CS: ICC = 0.58 ± 0.12 vs 0.51 ± 0.1) with 2DTagg. Finally, 3DST had similar global LS, but better global CS test-retest variability than 2DST. CONCLUSIONS:Using 2DTagg as reference, 3DST had better agreement and less bias for global and regional LS and CS. At the segmental level, 3DST demonstrated comparable agreement but lower bias versus 2DTagg compared with 2DST. Also, test-retest variability for global CS by 3DST was better than by 2DST. This suggests that 3DST is superior to 2DST for analysis of global and regional myocardial deformation, but further refinement is needed for both 3DST and 2DST at the segmental level.

The simulation of cardiac blood flow using patient-specific geometries can help for the diagnosis and treatment of cardiac diseases. Current patient-specific cardiac flow simulations requires a significant amount of human expertise and time to pre-process image data and obtain a case ready for simulations. A new procedure is proposed to alleviate this pre-processing by registering a unique generic mesh on patient-specific cardiac segmentations and transferring appropriately the spatiotemporal dynamics of the ventricle. The method is applied on real patient data acquired from 3D ultrasound imaging. Both a healthy and a pathological conditions are simulated. The resulting simulations exhibited physiological flow behavior in cardiac cavities. The experiments confirm a significant reduction in pre-processing work.

Background— Despite widespread use to characterize and refine prognosis, validation data of two-dimensional (2D) speckle tracking (2DST) echocardiography myocardial strain measurement remain scarce. Methods and Results— Global and regional subendocardial peak-systolic Lagrangian longitudinal (LS) and circumferential strain (CS) by 2DST and 2D-tagged (2DTagg) cardiac magnetic resonance imaging were compared against sonomicrometry in a dynamic heart phantom and among each other in 136 patients included prospectively at 2 centers. The ability of regional LS and CS 2DST and 2DTagg to identify late gadolinium enhancement was compared using receiver operating characteristics curves. In vitro, both LS-2DST and 2DTagg highly agreed with sonomicrometry (intraclass correlation coefficient [ICC], 0.89 and ICC, 0.90, both P<0.001 with −3±2.8% and 0.34±4.35% bias, respectively). In patients, both global LS and global CS 2DST agreed well with 2DTagg (ICC, 0.89 and ICC, 0.80; P<0.001); however, they provided systematically greater values (relative bias of −37±27% and −25±37% for global LS and global CS, respectively). On regional basis, however, ICC (from 0.17 to 0.81) and relative bias (from −9 to −98%) between 2DST and 2DTagg varied strongly among segments. Ability to discriminate infarcted versus noninfarcted segments by late gadolinium enhancement was similarly good for regional LS 2DTagg and 2DST (area under the curve, 0.66 versus 0.59; P=0.08), while it was lower for CS 2DST than 2DTagg (area under the curve, 0.61 versus 0.75; P<0.001). Conclusions— The high accuracy against sonomicrometry and good agreement of global LS and global CS by 2DST and 2DTagg confirm the overall validity of 2DST strain measurement. Yet, higher intertechnique segmental variability and lower ability for detecting infarct suggest that 2DST strain estimates may be less performant on regional than on global basis.

Diagnosing and localizing myocardial infarct is crucial for early patient management and therapy planning. We propose a new method for predicting the location of myocardial infarct from local wall deformation, which has value for risk stratification from routine examinations such as (3D) echocardiography. The pipeline combines non-linear dimensionality reduction of deformation patterns and two multi-scale kernel regressions. Confidence in the diagnosis is assessed by a map of local uncertainties, which integrates plausible infarct locations generated from the space of reduced dimensionality. These concepts were tested on 500 synthetic cases generated from a realistic cardiac electromechanical model, and 108 pairs of 3D echocardiographic sequences and delayed-enhancement magnetic resonance images from real cases. Infarct prediction is made at a spatial resolution around 4 mm, more than 10 times smaller than the current diagnosis, made regionally. Our method is accurate, and significantly outperforms the clinically-used thresholding of the deformation patterns (on real data: sensitivity/specificity of 0.828/0.804, area under the curve: 0.909 versus 0.742 for the most predictive strain component). Uncertainty adds value to refine the diagnosis and eventually re-examine suspicious cases.

This paper presents a novel algorithm that extends HARP to handle 3D tagged MRI images. HARP results were regularized by an original regularization framework defined in an anatomical space of coordinates. In the meantime, myocardium incompressibility was integrated in order to correct the radial strain which is reported to be more challenging to recover. Both the tracking and regularization of LV displacements were done on a volumetric mesh to be computationally efficient. Also, a window-weighted regression method was extended to cardiac motion tracking which helps maintain a low complexity even at finer scales. On healthy volunteers, the tracking accuracy was found to be as accurate as the best candidates of a recent benchmark. Strain accuracy was evaluated on synthetic data, showing low bias and strain errors under 5% (excluding outliers) for longitudinal and circumferential strains, while the second and third quartiles of the radial strain errors are in the ( 5%, 5%) range. In clinical data, strain dispersion was shown to correlate with the extent of transmural fibrosis. Also, reduced deformation values were found inside infarcted segments. (C) 2015 Elsevier B.V. All rights reserved.

BACKGROUND:Three-dimensional echocardiography (3DE) is a reliable and reproducible tool for assessing left ventricular (LV) function but remains sensitive to patient echogenicity. Contrast-enhanced 3DE (C3DE) has the potential to improve quantification in challenging patients. The aim of this study was to evaluate the impact of temporal resolution, spatial resolution, and image dynamic range on LV function assessed using C3DE compared with cardiac magnetic resonance imaging (MRI) in patients with poor echogenicity.METHODS:Forty-one patients with poor echogenicity who underwent two-dimensional echocardiography (2DE), 3DE, C3DE, and MRI were retrospectively investigated.RESULTS:Before contrast injection, 24 patients had three or more nonvisible segments. Three cases of 2DE and 12 cases of 3DE were not suitable for quantification. LV end-diastolic volumes were systematically underestimated by 2DE (142 ± 58 mL), 3DE (146 ± 69 mL), and C3DE (172 ± 61 mL) compared with MRI (216 ± 85 mL) (P < .001). Similar results were found for LV end-systolic volumes (81 ± 65 mL for 2DE, 82 ± 69 mL for 3DE, and 102 ± 80 mL for C3DE vs 129 ± 94 mL for MRI; P < .001). C3DE provided the best agreement with MRI (Lin concordance correlation coefficients of 0.67, 0.93, and 0.99, respectively, for end-diastolic volume, end-systolic volume, and ejection fraction) as well as the best measurement reproducibility. Finally, ultrasound settings had no significant effect on LV volumes and ejection fraction measurements.CONCLUSIONS:In these patients with poor ultrasound image quality, C3DE, regardless of instrument settings, outperformed 2DE and 3DE to assess LV volumes and ejection fraction and can thus be proposed as an acceptable alternative when MRI cannot be performed in this subgroup.

While abnormal patterns of cardiac electrophysiological activation are at the origin of important cardiovascular diseases (e.g., arrhythmia, asynchrony), the only clinically available method to observe detailed left ventricular endocardial surface activation pattern is through invasive catheter mapping. However, this electrophysiological activation controls the onset of the mechanical contraction; therefore, important information about the electrophysiology could be deduced from the detailed observation of the resulting motion patterns. In this paper, we present the study of this inverse cardiac electrokinematic relationship. The objective is to predict the activation pattern knowing the cardiac motion from the analysis of cardiac image sequences. To achieve this, we propose to create a rich patient-specific database of synthetic time series of the cardiac images using simulations of a personalized cardiac electromechanical model, in order to study this complex relationship between electrical activity and kinematic patterns in the context of this specific patient. We use this database to train a machine-learning algorithm which estimates the depolarization times of each cardiac segment from global and regional kinematic descriptors based on displacements or strains and their derivatives. Finally, we use this learning to estimate the patient’s electrical activation times using the acquired clinical images. Experiments on the inverse electrokinematic learning are demonstrated on synthetic sequences and are evaluated on clinical data with promising results. The error calculated between our prediction and the invasive intracardiac mapping ground truth is relatively small (around 10 ms for ischemic patients and 20 ms for nonischemic patient). This approach suggests the possibility of noninvasive electrophysiological pattern estimation using cardiac motion imaging.

This paper describes the data setup of the second cardiac Motion Analysis Challenge (cMac2). The purpose of this challenge is to initiate a public data repository for the benchmark of motion and strain quantification algorithms on 3D ultrasound images. The data currently includes synthetic images that combine ultrasound and biomechanical simulators. We also collected sonomicrometry curves and ultrasound images acquired on a Polyvinyl alcohol phantom.

This paper describes an algorithm for motion and deformation quantification of 3D cardiac ultrasound sequences. The algorithm is based on the assumption that the deformation field is smooth inside the myocardium. Thus, we assume that the displacement field can be represented as the convolution of an unknown field with a Gaussian kernel. We apply our algorithm to datasets with reliable ground truth: a set of synthetic sequences with known trajectories and a set of sequences of a mechanical phantom implanted with microsonometry crystals.

We propose a new approach for the generation of synthetic but visually realistic time series of cardiac images based on an electromechanical model of the heart and real clinical 4-D image sequences. This is achieved by combining three steps. The first step is the simulation of a cardiac motion using an electromechanical model of the heart and the segmentation of the end diastolic image of a cardiac sequence. We use biophysical parameters related to the desired condition of the simulated subject. The second step extracts the cardiac motion from the real sequence using nonrigid image registration. Finally, a synthetic time series of cardiac images corresponding to the simulated motion is generated in the third step by combining the motion estimated by image registration and the simulated one. With this approach, image processing algorithms can be evaluated as we know the ground-truth motion underlying the image sequence. Moreover, databases of visually realistic images of controls and patients can be generated for which the underlying cardiac motion and some biophysical parameters are known. Such databases can open new avenues for machine learning approaches.

In this paper, we propose to create a rich database of synthetic time series of 3D echocardiography (US) images using simulations of a cardiac electromechanical model, in order to study the relationship between electrical disorders and kinematic patterns visible in medical images. From a real 4D sequence, a software pipeline is applied to create several synthetic sequences by combining various steps including motion tracking and segmentation. We use here this synthetic database to train a machine learning algorithm which estimates the depolarization times of each cardiac segment from invariant kinematic descriptors such as local displacements or strains. First experiments on the inverse electrokinematic learning are demonstrated on the synthetic 3D US database and are evaluated on clinical 3D US sequences from two patients with Left Bundle Branch Block.

Despite advances in both medical image analysis and intracardiac electrophysiological mapping technology, the understanding of cardiac mechano-electrical coupling is still incomplete. This knowledge is of high interest since it would help estimating the cardiac electrophysiology function from the analysis of widely available cardiac images, such as 3D echocardiography. This is important, for example, in the evaluation of the cardiac resynchronization therapy (CRT) where the placement and tuning of the pacemaker leads plays a crucial role in the outcome of the therapy. This paper proposes a method to estimate activation times of myocardium using a cardiac electromechanical model. We use Kernel Ridge Regression to find the relationship between the kinematic descriptors (strain and displacement) and the contraction force caused by the action potential propagation. This regression model is then applied to two 3D echocardiographic sequences from a patient, one in sinus rhythm and the other one with left ventricle pacing, for which strains and displacements have been estimated using incompressible diffeomorphic demons for non-rigid registration.