High-Resolution Maps of Left Atrial Displacements and Strains Estimated with 3D Cine MRI using Online Learning Neural Networks

TL;DR

Results showed that Aladdin can accurately track the LA wall across the cardiac cycle and characterize its motion and deformation, and expects Aladdin to have important clinical applications as it can non-invasively characterize atrial pathophysiology.

Abstract

The functional analysis of the left atrium (LA) is important for evaluating cardiac health and understanding diseases like atrial fibrillation. Cine MRI is ideally placed for the detailed 3D characterization of LA motion and deformation but is lacking appropriate acquisition and analysis tools. Here, we propose tools for the Analysis for Left Atrial Displacements and DeformatIons using online learning neural Networks (Aladdin) and present a technical feasibility study on how Aladdin can characterize 3D LA function globally and regionally. Aladdin includes an online segmentation and image registration network, and a strain calculation pipeline tailored to the LA. We create maps of LA Displacement Vector Field (DVF) magnitude and LA principal strain values from images of 10 healthy volunteers and 8 patients with cardiovascular disease (CVD), of which 2 had large left ventricular ejection fraction (LVEF) impairment. We additionally create an atlas of these biomarkers using the data from the healthy volunteers. Results showed that Aladdin can accurately track the LA wall across the cardiac cycle and characterize its motion and deformation. Global LA function markers assessed with Aladdin agree well with estimates from 2D Cine MRI. A more marked active contraction phase was observed in the healthy cohort, while the CVD LVEF group showed overall reduced LA function. Aladdin is uniquely able to identify LA regions with abnormal deformation metrics that may indicate focal pathology. We expect Aladdin to have important clinical applications as it can non-invasively characterize atrial pathophysiology. All source code and data are available at: https://github.com/cgalaz01/aladdin_cmr_la.

Figures (5)

• Figure 1: Aladdin's image registration workflow. It consists of: 1) an online learning (subject-by-subject basis) nnU-Net that estimates the LA segmentation maps, 2) contours from the LA segmentation maps are first extracted and then dilated to create masks for the images, and 3) an online learning 3D U-Net-like image registration network, Aladdin-R.
• Figure 2: The spatially averaged Dice score (orange line) and Hausdorff distance (mm) (blue dotted line) for the online learning nnU-Net across the cardiac cycle when trained on cardiac phases 0, 8, and 15 across all 18 subjects. The shaded areas around each line, colored to match the respective mean, indicate the standard deviation.
• Figure 3: Comparison between healthy (green), CVD (brown), and CVD $\text{LVEF}_\downarrow$ (red) groups across the cardiac cycle and their respective t-test $p$-values between healthy/CVD (dotted brown) and healthy/CVD $\text{LVEF}_\downarrow$ (dotted red). Plot A shows the DVF magnitude and plot B the first Green-Lagrangian principal strain values. The values shown are after registering each case to the atlas to reduce variance from LA size differences.
• Figure 4: Atlas of the LA Displacements and Principal Strains: From left to right: 1) Whole heart view with the LA atlas, 2) the generated DVF, and 2) the estimated first principal strains for the anterior view at the end-reservoir phase (phase 8). The atlas across the cardiac cycle can be viewed in SupVid https://github.com/cgalaz01/aladdin_cmr_la/blob/main/supplements/README.md#SupVid9
• Figure 5: Mahalanobis distance of the first principal strain between the atlas distribution and the estimates for three representative patients with, from left to right: myocarditis, myocardial infarction (CVD $\text{LVEF}_\downarrow$) and non-ischaemic cardiomyopathy. The anterior view at the start of the boost-pump phase is shown. Regions with a high Mahalanobis distance are highlighted with yellow boxes, which may indicate areas of regional functional abnormality.