Learning-Based Regularization for Cardiac Strain Analysis via Domain Adaptation

Abstract

Reliable motion estimation and strain analysis using 3D+time echocardiography (4DE) for localization and characterization of myocardial injury is valuable for early detection and targeted interventions. However, motion estimation is difficult due to the low-SNR that stems from the inherent image properties of 4DE, and intelligent regularization is critical for producing reliable motion estimates. In this work, we incorporated the notion of domain adaptation into a supervised neural network regularization framework. We first propose a semi-supervised Multi-Layered Perceptron (MLP) network with biomechanical constraints for learning a latent representation that is shown to have more physiologically plausible displacements. We extended this framework to include a supervised loss term on synthetic data and showed the effects of biomechanical constraints on the network's ability for domain adaptation. We validated the semi-supervised regularization method on in vivo data with implanted sonomicrometers. Finally, we showed the ability of our semi-supervised learning regularization approach to identify infarct regions using estimated regional strain maps with good agreement to manually traced infarct regions from postmortem excised hearts.

A Novel 3D-to-3D Diffeomorphic Registration Algorithm With Applications to Left Ventricle Segmentation in MR and Ultrasound Sequences

“…There have been several approaches proposed in the literature for where 3D registration of the LV over a temporal sequence is used for computing the deformation fields [21] for US sequences. These can be divided into areas including intensitybased, regularization model-based approaches, and also deep learning methods.…”

“…A deep learning technique for performing both motion and strain estimation has been developed by [21]. The authors first use a multi-layer perceptron in a supervised manner in order to learn the features between the input and ground truth deformation fields.…”

Deep learning in ultrasound elastography imaging: A review

“…This work showed the capability of identifying infarcted cardiac regions in a canine model, which was verified by manually traced infarcted regions from postmortem excised hearts. 143…”

“…A more recent work extended the multilayered perceptron network to include biomechanical constraints. 143 Incompressibility and periodic motion of cardiac tissues were introduced into the cost function. The biomechanically inspired constraints alleviated the poor tracking F I G U R E 6 Overview of classic neural network methods and the learning-using-privileged-information (LUPI) network.…”

Novel Cardiac Computed Tomography Methods for the Assessment of Anthracycline Induced Cardiotoxicity

“…In patients undergoing transcatheter aortic valve replacement CT strain was predictive of adverse outcomes, ( 36 ) and improvement in CT derived global longitudinal and principal strain has been demonstrated after transcatheter aortic valve replacement ( 34 ). The optimal evaluation of regional strain from CT images should employ 3-dimensional tracking of regional myocardial displacements ( 37 ), and probably will employ deep learning ( 39 ). Automated segmentation of left ventricular cavity in temporal cardiac image sequences (consisting of multiple time-points) is a fundamental requirement for quantitative analysis of cardiac structural and functional changes.…”