Iterated Residual Graph Convolutional Neural Network for Personalized Three-Dimensional Reconstruction of Left...

Area of Science:

• Medical imaging
• Computational anatomy
• Cardiovascular research

Background:

• Accurate three-dimensional (3D) reconstruction of the left myocardium is crucial for diagnosing and treating cardiac diseases.
• Existing model-based reconstruction methods are limited by the similarity between the target object and average models, impacting mesh accuracy.

Purpose of the Study:

• To develop a personalized 3D reconstruction algorithm for the left myocardium utilizing cardiac Magnetic Resonance (MR) images.
• To enhance the accuracy and efficiency of 3D myocardial reconstruction through a novel deep learning approach.

Main Methods:

• A residual graph convolutional neural network (GCN) was employed for mesh deformation of the left myocardium.
• An initial triangular mesh was generated from segmentation results and subsequently refined using the iterated residual GCN.
• A vertex feature learning module, using an encoder-decoder network, was incorporated to capture myocardial skeletal information at various receptive fields, guiding mesh deformation.

Main Results:

• The proposed algorithm successfully reconstructed the 3D shape of the left myocardium from cardiac MR images.
• Comparative experiments demonstrated that the developed method achieves competitive and accurate results compared to existing state-of-the-art approaches.
• The integration of shape and local relationships significantly improved the mesh deformation process.

Conclusions:

• The proposed personalized 3D reconstruction algorithm based on residual GCN offers a robust and accurate method for left myocardial reconstruction from cardiac MR images.
• This approach holds significant potential for improving the diagnosis and treatment planning of cardiac diseases.
• The vertex feature learning module effectively utilizes anatomical information to enhance reconstruction fidelity.