DeepStrain: A Deep Learning Workflow for the Automated Characterization of Cardiac Mechanics

Abstract

Myocardial strain analysis from cinematic magnetic resonance imaging (cine-MRI) data could provide a more thorough characterization of cardiac mechanics than volumetric parameters such as left-ventricular ejection fraction, but sources of variation including segmentation and motion estimation have limited its wide clinical use. We designed and validated a deep learning (DL) workflow to generate both volumetric parameters and strain measures from cine-MRI data, including strain rate (SR) and regional strain polar maps, consisting of segmentation and motion estimation convolutional neural networks developed and trained using healthy and cardiovascular disease (CVD) subjects (n=150). DL-based volumetric parameters were correlated (>0.98) and without significant bias relative to parameters derived from manual segmentations in 50 healthy and CVD subjects. Compared to landmarks manually-tracked on tagging-MRI images from 15 healthy subjects, landmark deformation using DL-based motion estimates from paired cine-MRI data resulted in an end-point-error of 2.9 {+/-} 1.5 mm. Measures of end-systolic global strain from these cine-MRI data showed no significant biases relative to a tagging-MRI reference method. On 4 healthy subjects, intraclass correlation coefficient for intra-scanner repeatability was excellent (>0.95) for strain, moderate to excellent for SR (0.690-0.963), and good to excellent (0.826-0.994) in most polar map segments. Absolute relative change was within ~5% for strain, within ~10% for SR, and <1% in half of polar map segments. In conclusion, we developed and evaluated a DL-based, end-to-end fully-automatic workflow for global and regional myocardial strain analysis to quantitatively characterize cardiac mechanics of healthy and CVD subjects based on ubiquitously acquired cine-MRI data.