IGCN+/IGCN is a learning framework for 2D/3D deformable model registration and alignment, and shape reconstruction from a single-viewpoint projection image. The generative network learns translation from the input projection image to a displacement map, and the GCN learns mesh deformation based on the sampled per-vertex feature and connectivity.
IGCN_movie.mp4
- Left (input): digitally reconstructed radiograph images generated from 4D-CT data (10-frame sequential volumes)
- Center (output): registered mesh of abdominal organs
- Right (error): target (magenta) and predicted (cyan) mesh
- Python 3.9
- NVIDIA CUDA 11.2.0 and cuDNN 8.1.1
- TFLearn with Tensorflow backend
If you use this code for your research, please cite
-
IGCN+ (The latest version):
M. Nakao, M. Nakamura, T. Matsuda, Image-to-graph convolutional network for 2D/3D deformable model registration of low-contrast organs, IEEE Trans. on Medical Imaging, Vol. 41, No. 12, pp. 3747-3761, 2022. https://ieeexplore.ieee.org/document/9844010 -
IGCN (MICCAI version):
M. Nakao, M. Nakamura, T. Matsuda, Image-to-Graph Convolutional Network for Deformable Shape Reconstruction from a Single Projection Image, International Conference on Medical Image Computing and Computer Assisted Intervention (MICCAI), pp. 259-268, 2021. https://link.springer.com/chapter/10.1007/978-3-030-87202-1_25