SHAPE-REGULARIZED UNSUPERVISED LEFT VENTRICULAR MOTION NETWORK WITH SEGMENTATION CAPABILITY IN 3D+TIME ECHOCARDIOGRAPHY

Abstract

Accurate motion estimation and segmentation of the left ventricle from medical images are important tasks for quantitative evaluation of cardiovascular health. Echocardiography offers a cost-efficient and non-invasive modality for examining the heart, but provides additional challenges for automated analyses due to the low signal-to-noise ratio inherent in ultrasound imaging. In this work, we propose a shape regularized convolutional neural network for estimating dense displacement fields between sequential 3D B-mode echocardiography images with the capability of also predicting left ventricular segmentation masks. Manually traced segmentations are used as a guide to assist in the unsupervised estimation of displacement between a source and a target image while also serving as labels to train the network to additionally predict segmentations. To enforce realistic cardiac motion patterns, a flow incompressibility term is also incorporated to penalize divergence. Our proposed network is evaluated on an in vivo canine 3D+t B-mode echocardiographic dataset. It is shown that the shape regularizer improves the motion estimation performance of the network and our overall model performs favorably against competing methods.

Keywords: 3D+t echocardiography; deep learning; motion tracking; segmentation.

Fig. 1.

Proposed motion tracking network. Motion estimation is trained in an unsupervised manner with the assistance of manually traced segmentations acting as a shape consistency constraint and can be used to simultaneously generate segmentation predictions. Green boxes indicate model output.

Fig. 2.

Short-axis view of motion vectors between paired time frames (top) and segmentations (bottom - green: manual tracing, red: prediction) on a baseline canine echocardiographic sequence. A) Free-form Deformation, B) Optical Flow, C) Proposed Model, D) Proposed Model + Regularization