Ventricle Surface Reconstruction from Cardiac MR Slices Using Deep Learning

Xu, Hao; Zacur, Ernesto; Schneider, Jürgen; Grau, Vicente

doi:10.1007/978-3-030-21949-9_37

Cited by 11 publications

(9 citation statements)

References 26 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The point cloud branch architectures ( Figure 2A,B ) are able to apply deep learning operations directly on point cloud data, which allows surface data of much higher resolution to be efficiently processed and used for storing anatomical shape information. This is in contrast to the widely-used voxelgrid representations ( Çiçek et al, 2016 ; Bello et al, 2019 ; Xu et al, 2019 ), which are considerably less memory-efficient at managing surface-level data leading to lower resolution, longer processing times, and ultimately limit the overall accuracy of the modeled anatomy. Furthermore, each of the high-dimensional point clouds combines both the left and right ventricular anatomy and maintains separate labels for the LV endocardium, LV epicardium, and RV endocardium substructures.…”

Section: Discussionmentioning

confidence: 94%

Multi-Domain Variational Autoencoders for Combined Modeling of MRI-Based Biventricular Anatomy and ECG-Based Cardiac Electrophysiology

2022

View full text Add to dashboard Cite

Human cardiac function is characterized by a complex interplay of mechanical deformation and electrophysiological conduction. Similar to the underlying cardiac anatomy, these interconnected physiological patterns vary considerably across the human population with important implications for the effectiveness of clinical decision-making and the accuracy of computerized heart models. While many previous works have investigated this variability separately for either cardiac anatomy or physiology, this work aims to combine both aspects in a single data-driven approach and capture their intricate interdependencies in a multi-domain setting. To this end, we propose a novel multi-domain Variational Autoencoder (VAE) network to capture combined Electrocardiogram (ECG) and Magnetic Resonance Imaging (MRI)-based 3D anatomy information in a single model. Each VAE branch is specifically designed to address the particular challenges of the respective input domain, enabling efficient encoding, reconstruction, and synthesis of multi-domain cardiac signals. Our method achieves high reconstruction accuracy on a United Kingdom Biobank dataset, with Chamfer Distances between reconstructed and input anatomies below the underlying image resolution and ECG reconstructions outperforming multiple single-domain benchmarks by a considerable margin. The proposed VAE is capable of generating realistic virtual populations of arbitrary size with good alignment in clinical metrics between the synthesized and gold standard anatomies and Maximum Mean Discrepancy (MMD) scores of generated ECGs below those of comparable single-domain approaches. Furthermore, we observe the latent space of our VAE to be highly interpretable with separate components encoding different aspects of anatomical and ECG variability. Finally, we demonstrate that the combined anatomy and ECG representation improves the performance in a cardiac disease classification task by 3.9% in terms of Area Under the Receiver Operating Characteristic (AUROC) curve over the best corresponding single-domain modeling approach.

show abstract

Section: Discussionmentioning

confidence: 94%

Multi-Domain Variational Autoencoders for Combined Modeling of MRI-Based Biventricular Anatomy and ECG-Based Cardiac Electrophysiology

2022

View full text Add to dashboard Cite

show abstract

“…voxelgrids) required by classical deep learning approaches (e.g. [11]), which are inefficient at storing sparse surface data and have limited resolution leading to partial volume effects. This reduced spatial efficiency coupled with longer running times is especially disadvantageous for applications in clinical practice and large-scale cardiac physiology simulations.…”

Section: Discussionmentioning

confidence: 99%

“…We sample the translation and rotation values for each level from separate normal dis-tributions with zero mean. The respective standard deviations are selected to reflect realistic values observed in clinical practice [11] (translation: 1.5 mm, 2.5 mm, 3.5 mm; rotation: 0.5 • , 1.5 • , 2.5 • ). Finally, we slice the mesh at the randomly misaligned planes and sample points along each resulting contour to create the input point clouds.…”

Section: Training Data Generationmentioning

confidence: 99%

“…To the best of our knowledge, this is the first application of geometric deep learning to cardiac surface reconstruction from cine MRI contours and the first deep learning method overall to demonstrate successful biventricular surface reconstruction on real cine MR images. Previous deep learning approaches either focused only on the left ventricle [4], a different modality [10], or lacked validation on real data [11]. Fig.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Biventricular Surface Reconstruction From Cine Mri Contours Using Point Completion Networks

Beetz

Banerjee

Grau

2021

2021 IEEE 18th International Symposium on Biomedical Imaging (ISBI)

Self Cite

View full text Add to dashboard Cite

Many important cardiac biomarkers used in clinical practice describe cardiac anatomy and function in three dimensions (3D). However, common cardiac magnetic resonance imaging (MRI) protocols often only generate two-dimensional (2D) image slices of the underlying 3D anatomy and are susceptible to various types of motion artifacts causing slice misalignment. In this paper, we propose a deep learning method acting directly on point clouds to reconstruct a dense 3D biventricular heart model from misaligned 2D cardiac MR image contours. The method is able to reduce mild, medium, and strong slice misalignments (mean translation ∼3.5 mm; mean rotation ∼2.5 • ) to a Chamfer distance below image resolution (1.25 mm) with high robustness (standard deviation 0.18 mm) on a statistical shape model dataset. It also manages to reconstruct smooth 3D shapes with accurate left ventricular volumes from cine MR images of the UK Biobank study.

show abstract

“…A method for simultaneous misalignment and segmentation correction for cardiac images with a final 3D surface reconstruction was introduced in [ 22 ]. A deep learning-based method was proposed in [ 23 ], where the problem of mesh fitting from sparse inputs was transformed into a 3D volumetric inpainting problem followed by isosurfacing from dense volumetric data.…”

Section: Introductionmentioning

confidence: 99%

A completely automated pipeline for 3D reconstruction of human heart from 2D cine magnetic resonance slices

Banerjee

Camps

Zacur

et al. 2021

Phil. Trans. R. Soc. A.

Self Cite

View full text Add to dashboard Cite

Cardiac magnetic resonance (CMR) imaging is a valuable modality in the diagnosis and characterization of cardiovascular diseases, since it can identify abnormalities in structure and function of the myocardium non-invasively and without the need for ionizing radiation. However, in clinical practice, it is commonly acquired as a collection of separated and independent 2D image planes, which limits its accuracy in 3D analysis. This paper presents a completely automated pipeline for generating patient-specific 3D biventricular heart models from cine magnetic resonance (MR) slices. Our pipeline automatically selects the relevant cine MR images, segments them using a deep learning-based method to extract the heart contours, and aligns the contours in 3D space correcting possible misalignments due to breathing or subject motion first using the intensity and contours information from the cine data and next with the help of a statistical shape model. Finally, the sparse 3D representation of the contours is used to generate a smooth 3D biventricular mesh. The computational pipeline is applied and evaluated in a CMR dataset of 20 healthy subjects. Our results show an average reduction of misalignment artefacts from 1.82 ± 1.60 mm to 0.72 ± 0.73 mm over 20 subjects, in terms of distance from the final reconstructed mesh. The high-resolution 3D biventricular meshes obtained with our computational pipeline are used for simulations of electrical activation patterns, showing agreement with non-invasive electrocardiographic imaging. The automatic methodologies presented here for patient-specific MR imaging-based 3D biventricular representations contribute to the efficient realization of precision medicine, enabling the enhanced interpretability of clinical data, the digital twin vision through patient-specific image-based modelling and simulation, and augmented reality applications. This article is part of the theme issue ‘Advanced computation in cardiovascular physiology: new challenges and opportunities’.

show abstract

Ventricle Surface Reconstruction from Cardiac MR Slices Using Deep Learning

Cited by 11 publications

References 26 publications

Multi-Domain Variational Autoencoders for Combined Modeling of MRI-Based Biventricular Anatomy and ECG-Based Cardiac Electrophysiology

Multi-Domain Variational Autoencoders for Combined Modeling of MRI-Based Biventricular Anatomy and ECG-Based Cardiac Electrophysiology

Biventricular Surface Reconstruction From Cine Mri Contours Using Point Completion Networks

A completely automated pipeline for 3D reconstruction of human heart from 2D cine magnetic resonance slices

Contact Info

Product

Resources

About