Search for: All records

This study explores the application of deep learning to the segmentation of DENSE cardiovascular magnetic resonance (CMR) images, which is an important step in the analysis of cardiac deformation and may help in the diagnosis of heart conditions. A self-adapting method based on the nnU-Net framework is introduced to enhance the accuracy of DENSE-MR image segmentation, with a particular focus on the left ventricle myocardium (LVM) and left ventricle cavity (LVC), by leveraging the phase information in the cine DENSE-MR images. Two models are built and compared: 1) ModelM, which uses only the magnitude of the DENSE-MR images; and 2) ModelMP, which incorporates magnitude and phase images. DENSE-MR images from 10 human volunteers processed using the DENSE-Analysis MATLAB toolbox were included in this study. The two models were trained using a 2D UNet-based architecture with a loss function combining the Dice similarity coefficient (DSC) and cross-entropy. The findings show the effectiveness of leveraging the phase information with ModelMP resulting in a higher DSC and improved image segmentation, especially in challenging cases, e.g., at early systole and with basal and apical slices. 

Porcine hearts (N = 14) underwent ex vivo diffusion tensor imaging (DTI) at 3T. DTI analysis showed regional differences in helix angle (HA) range. The HA range in the posterior free wall was significantly greater than that of the anterior free wall (p = 0.02), the lateral free wall (p < 0.001) and the septum (p = 0.008). The best-fit transmural HA function also varied by region, with eight regions best described by an arctan function, seven by an arcsine function, and a single region by a linear function. Tractography analysis was performed, and the length that the tracts spanned within the epicardial, midwall, and endocardial segments was measured. A high number of tracts span the epicardial and mid-wall thirds, with fewer tracts spanning the mid-wall and endocardial thirds. Connectivity analysis of the number of tracts connecting different ventricular regions showed a high prevalence of oblique tracts that may be critical for long-range connectivity. 

In this work we present a machine learning model to segment long axis magnetic resonance images of the left ventricle (LV) and address the challenges encountered when, in doing so, a small training dataset is used. Our approach is based on a heart locator and an anatomically guided UNet model in which the UNet is followed by a B-Spline head to condition training. The model is developed using transfer learning, which enabled the training and testing of the proposed strategy from a small swine dataset. The segmented LV cavity and myocardium in the long axis view show good agreement with ground truth segmentations at different cardiac phases based on the Dice similarity coefficient. In addition the model provides a measure of segmentations' uncertainty, which can then be incorporated while developing LV computational models and indices of cardiac performance based on the segmented images. Finally, several challenges related to long axis, as opposed to short axis, image segmentation are highlighted, including proposed solutions.