Implement 4D support - Githubissues

kbressem commented 3 years ago

Create class to automatically matches spacing, resolution and orientation of multiple 3D images.

review-notebook-app[bot] commented 3 years ago

Check out this pull request on

See visual diffs & provide feedback on Jupyter Notebooks.

Powered by ReviewNB

kbressem commented 3 years ago

Motivation Most 3D radiology data consists out of multiple sequences and partial information from all sequences is needed to make the final diagnosis. Models should be able to process multiple 3D sequences (4D input) to have access to all the information. Technical assistants sometimes vary the angulation, direction or FOV of an sequence, especially in MRI. This mean, simply stacking 3D volumes on top of each other, is not a good solution, as this will lead to misplacement of voxels in the physical space.

proposed solution

a simple image registration tool which unifies the spacing, size, origin and direction of multiple TensorDicom3D images and TensorMask3D images.
modifications on the dataloaders to process multiple inputs
extension for 3D CNN and 3D Dynamic U-Net to accept multiple inputs.

kbressem / faimed3d

Implement 4D support #24