Correction of Slice Misalignment in Multi-breath-hold Cardiac MRI Scans

Abstract

Cardiac Magnetic Resonance (CMR) provides unique functional and anatomical visualisation of the macro and micro-structures of the heart. However, CMR acquisition times usually necessitate slices to be acquired at different breath holds, which results in potential misalignment of the acquired slices. Correcting for this spatial misalignment is required for accurate three-dimensional (3D) reconstruction of the heart chambers allowing robust metrics for shape analysis among populations as well as precise representations of individual geometries and scars. While several methods have been proposed to realign slices, their use in other important protocols such as late gadolinium enhancement ( LGE) is yet to be demonstrated. We propose a registration framework based on local phase to correct for slice misalignment. Our registration framework is a group registration technique combining long-and short-axis slices. Validation was performed on LGE slices using expert-traced ventricular contours. For 15 clinical multi-breath-hold datasets our method reduced the median discrepancy of moderately misaligned slices from 2.19 mm to 1.63 mm, and of severely misaligned from 7.33 mm to 1.96 mm.

MRI-Based Computational Torso/Biventricular Multiscale Models to Investigate the Impact of Anatomical Variability on the ECG QRS Complex

“…As the image resolution in standard MRI acquisitions does not allow to differentiate right ventricular epicardial and endocardial contours in the right ventricle, we synthesized right epicardial contours by a 3.5 mm offset from the endocardial contours (Prakash, 1978). Spatial misalignments in slice images and spatial discrepancies between the contours due to acquisitions at different breath holds were corrected by aligning intensity profiles of intersecting slices using a 3D rigid transformation for each image (Villard et al, 2017) (see Figure 1A, Contours alignment). Bi-ventricular geometries were built from the aligned contours using the end-diastole frames from the standard CINE acquisition as in Villard et al (2018) and Zacur et al (2017).…”

Biventricular Surface Reconstruction From Cine Mri Contours Using Point Completion Networks

“…Considerable research has focused on correcting slice misalignment induced by respiratory motion between breath holds [5,6] and on reconstructing 3D surfaces from sparse and noisy input data [7,8]. In this paper, we propose a fast and fully automatic geometric deep learning method, capable of addressing both the sparsity and misalignment problem in a single model.…”

Ventricle Surface Reconstruction from Cardiac MR Slices Using Deep Learning

“…A standard clinical CMR study includes a stack of short-axis (SAX) slices, covering at least from the left/right ventricular (LV/RV) apex to the base, plus at least two long-axis (LAX) views: horizontal long-axis (HLA, also known as 4 chamber view or 4CH) and vertical long-axis (VLA, also known as 2 chamber view or 2CH) [7]. Classical isosurfacing algorithms cannot be directly used due to the sparsity of the input data and because of the presence of motion artefacts (misalignment between slices caused by multiple breath holding and possible body movement during acquisition) [8], which make the task of reconstructing 3D structure from CMR data particularly challenging. Reconstruction of 3D surfaces from CMR data is normally formulated as a 3D mesh adaptation problem to sparse contours or points [9,10,11,12], and solutions often incorporate shape priors during that process.…”