We propose a method based on deep learning to perform cardiac segmentation on short axis Magnetic resonance imaging stacks iteratively from the top slice (around the base) to the bottom slice (around the apex). At each iteration, a novel variant of the U-net is applied to propagate the segmentation of a slice to the adjacent slice below it. In other words, the prediction of a segmentation of a slice is dependent upon the already existing segmentation of an adjacent slice. The 3-D consistency is hence explicitly enforced. The method is trained on a large database of 3078 cases from the U.K. Biobank. It is then tested on the 756 different cases from the U.K. Biobank and three other state-of-the-art cohorts (ACDC with 100 cases, Sunnybrook with 30 cases, and RVSC with 16 cases). Results comparable or even better than the state of the art in terms of distance measures are achieved. They also emphasize the assets of our method, namely, enhanced spatial consistency (currently neither considered nor achieved by the state of the art), and the generalization ability to unseen cases even from other databases.

This paper presents a new registration algorithm, called Temporal Diffeomorphic Free Form Deformation (TDFFD), and its application to motion and strain quantification from a sequence of 3D ultrasound (US) images. The originality of our approach resides in enforcing time consistency by representing the 4D velocity field as the sum of continuous spatiotemporal B-Spline kernels. The spatiotemporal displacement field is then recovered through forward Eulerian integration of the non-stationary velocity field. The strain tensor is computed locally using the spatial derivatives of the reconstructed displacement field. The energy functional considered in this paper weighs two terms: the image similarity and a regularization term. The image similarity metric is the sum of squared differences between the intensities of each frame and a reference one. Any frame in the sequence can be chosen as reference. The regularization term is based on the incompressibility of myocardial tissue. TDFFD was compared to pairwise 3D FFD and 3D+t FFD, both on displacement and velocity fields, on a set of synthetic 3D US images with different noise levels. TDFFD showed increased robustness to noise compared to these two state-of-the-art algorithms. TDFFD also proved to be more resistant to a reduced temporal resolution when decimating this synthetic sequence. Finally, this synthetic dataset was used to determine optimal settings of the TDFFD algorithm. Subsequently, TDFFD was applied to a database of cardiac 3D US images of the left ventricle acquired from 9 healthy volunteers and 13 patients treated by Cardiac Resynchronization Therapy (CRT). On healthy cases, uniform strain patterns were observed over all myocardial segments, as physiologically expected. On all CRT patients, the improvement in synchrony of regional longitudinal strain correlated with CRT clinical outcome as quantified by the reduction of end-systolic left ventricular volume at follow-up (6 and 12months), showing the potential of the proposed algorithm for the assessment of CRT.

Background: Current diagnosis of heart failure with preserved ejection fraction (HFpEF) is suboptimal. We tested the hypothesis that comprehensive machine learning (ML) of left ventricular function at rest and exercise objectively captures differences between HFpEF and healthy subjects. Methods and Results: One hundred fifty-six subjects aged >60 years (72 HFpEF+33 healthy for the initial analyses; 24 hypertensive+27 breathless for independent evaluation) underwent stress echocardiography, in the MEDIA study (Metabolic Road to Diastolic Heart Failure). Left ventricular long-axis myocardial velocity patterns were analyzed using an unsupervised ML algorithm that orders subjects according to their similarity, allowing exploration of the main trends in velocity patterns. ML identified a continuum from health to disease, including a transition zone associated to an uncertain diagnosis. Clinical validation was performed (1) to characterize the main trends in the patterns for each zone, which corresponded to known characteristics and new features of HFpEF; the ML-diagnostic zones differed for age, body mass index, 6-minute walk distance, B-type natriuretic peptide, and left ventricular mass index ( P <0.05) and (2) to evaluate the consistency of the proposed groupings against diagnosis by current clinical criteria; correlation with diagnosis was good (κ, 72.6%; 95% confidence interval, 58.1–87.0); ML identified 6% of healthy controls as HFpEF. Blinded reinterpretation of imaging from subjects with discordant clinical and ML diagnoses revealed abnormalities not included in diagnostic criteria. The algorithm was applied independently to another 51 subjects, classifying 33% of hypertensive and 67% of breathless controls as mild-HFpEF. Conclusions: The analysis of left ventricular long-axis function on exercise by interpretable ML may improve the diagnosis and understanding of HFpEF.

Convolutional neural networks (CNN) have demonstrated their ability to segment 2D cardiac ultrasound images. However, despite recent successes according to which the intra-observer variability on end-diastole and end-systole images has been reached, CNNs still struggle to leverage temporal information to provide accurate and temporally consistent segmentation maps across the whole cycle. Such consistency is required to accurately describe the cardiac function, a necessary step in diagnosing many cardiovascular diseases. In this paper, we propose a framework to learn the 2D+time apical long-axis cardiac shape such that the segmented sequences can benefit from temporal and anatomical consistency constraints. Our method is a post-processing that takes as input segmented echocardiographic sequences produced by any state-of-the-art method and processes it in two steps to (i) identify spatio-temporal inconsistencies according to the overall dynamics of the cardiac sequence and (ii) correct the inconsistencies. The identification and correction of cardiac inconsistencies relies on a constrained autoencoder trained to learn a physiologically interpretable embedding of cardiac shapes, where we can both detect and fix anomalies. We tested our framework on 98 full-cycle sequences from the CAMUS dataset, which are available alongside this paper. Our temporal regularization method not only improves the accuracy of the segmentation across the whole sequences, but also enforces temporal and anatomical consistency.

Aims Pathophysiology of heart failure (HF) with preserved ejection fraction (HFPEF) remains unclear. Left atrial (LA) function has been related to HF symptoms. Our purpose is to analyse LA function in outpatients with new onset symptoms of HF. Methods and results An observational study was performed including 138 consecutive outpatients with suspected HF referred to a one-stop clinic. Final diagnosis [HF with reduced EF (HFREF), HFPEF, or non-HF] was established according to current recommendations. Echocardiography was performed in all patients. LA function was analysed using strain derived from speckle tracking in sinus rhythm patients ( n = 83). Results were analysed with ANOVA and Bonferroni statistical tests. Receiver operating characteristic (ROC) curves were constructed to investigate the predictive ability of LA parameters for the final diagnosis of HF. Patients were 75 ± 9 years and 63% women. Final diagnosis was 23.2% HFREF, 45.7% HFPEF, and 31.2% non-HF. Left ventricular strain rate showed no differences between non-HF and HFPEF groups, but both groups showed differences with the HFREF group. LA strain rate (A- and S-waves) was significantly reduced in both HF groups (without differences among them) when compared with the non-HF group. LA strain rate and indexed volume showed significant accuracy for HF diagnosis in ROC curves. Conclusions In outpatients with new-onset symptoms of HF, LA dysfunction was observed. It might be the initial mechanism in the development of symptoms in HFPEF patients. These findings support the relationship of LA dysfunction with HFPEF, suggesting that the analysis of LA function may be useful in sinus rhythm patients with new-onset dyspnoea.

Atlases and statistical models play important roles in the personalization and simulation of cardiac physiology. For the study of the heart, however, the construction of comprehensive atlases and spatio-temporal models is faced with a number of challenges, in particular the need to handle large and highly variable image datasets, the multi-region nature of the heart, and the presence of complex as well as small cardiovascular structures. In this paper, we present a detailed atlas and spatio-temporal statistical model of the human heart based on a large population of 3D+time multi-slice computed tomography sequences, and the framework for its construction. It uses spatial normalization based on nonrigid image registration to synthesize a population mean image and establish the spatial relationships between the mean and the subjects in the population. Temporal image registration is then applied to resolve each subject-specific cardiac motion and the resulting transformations are used to warp a surface mesh representation of the atlas to fit the images of the remaining cardiac phases in each subject. Subsequently, we demonstrate the construction of a spatio-temporal statistical model of shape such that the inter-subject and dynamic sources of variation are suitably separated. The framework is applied to a 3D+time data set of 138 subjects. The data is drawn from a variety of pathologies, which benefits its generalization to new subjects and physiological studies. The obtained level of detail and the extendability of the atlas present an advantage over most cardiac models published previously.