Transformer-Guided Graph Attention for Direct Cardiac Mesh Reconstruction: A Structural Digital Twin Framework

The article presents a novel end-to-end network for generating patient-specific cardiac meshes directly from 3D medical images, utilizing a 3D Swin Transformer encoder-decoder combined with a Graph Attention Network (GAT) head. This method achieves competitive segmentation scores (Dice of 0.84 for CT and 0.83 for MRI) and significantly improves mesh quality with a mean Chamfer distance of 1.8 mm, while eliminating the need for traditional mesh generation techniques like Marching Cubes. This advancement enhances the accessibility of cardiac simulations in clinical settings by streamlining the workflow and ensuring high geometric fidelity and topological correctness.