Explicit differentiable slicing and global deformation for cardiac mesh reconstruction

Area of Science:

• Medical imaging
• Computational anatomy
• Biophysics

Background:

• 3D cardiac mesh reconstruction from medical images is vital for analysis but challenging due to sparse, noisy 2D slices.
• Traditional methods lack fidelity or require extensive 3D annotations.
• Differentiable supervision from 2D to 3D is needed for end-to-end optimization.

Purpose of the Study:

• To develop a novel differentiable algorithm for 3D mesh reconstruction from 2D medical images.
• To create a framework for patient-specific left ventricle (LV) mesh extraction using this algorithm.
• To enable accurate cardiac shape and motion analysis from sparse imaging data.

Main Methods:

• Proposed an explicit differentiable voxelization and slicing (DVS) algorithm for gradient backpropagation from 2D slices to 3D meshes.
• Developed a framework coupling DVS with graph harmonic deformation (GHD) for cardiac shape morphing.
• Utilized losses defined on 2D images for direct supervision of 3D mesh optimization.

Main Results:

• Achieved state-of-the-art performance in cardiac mesh reconstruction from both dense (CT) and sparse (MRI) images.
• Demonstrated an overall Dice score of 90% in sparse fitting across multiple datasets.
• Accurately quantified clinical parameters like ejection fraction and global myocardial strains, outperforming traditional methods on sparse data.

Conclusions:

• The proposed DVS algorithm and GHD framework enable accurate and efficient 3D cardiac mesh reconstruction from sparse 2D medical images.
• This method overcomes limitations of traditional approaches, offering improved fidelity and reduced annotation requirements.
• The framework facilitates precise quantification of cardiac function, advancing cardiovascular research and clinical applications.