search

NILOIDE / CMR_intensity_3Dmoco

We propose a method to mitigate the effect of inter-slice motion in CMR images for all SA and LA slices simultaneously by optimizing the 3D rotation and translation parameters on sampled intensities along slice intersections.
2 stars 0 forks source link

readme

Intensity-based 3D motion correction for cardiac MR images

This is the code repository for our ISBI submission:

We propose a method to mitigate the effect of inter-slice motion in CMR images for all SA and LA slices simultaneously by optimizing the 3D rotation and translation parameters on sampled intensities along slice intersections. Our approach is formulated as a subject-specific optimization problem and requires no prior knowledge of the underlying anatomy.

Alt text

Sample results

If two slices are aligned, both slices should have equal intensities along their intersections. We observe clear similarities in intersection intensity patterns (vertical bands) after optimization (green lines) as opposed to the starting alignment (red lines).

Alt text

We sample points along the intersection and minimize the difference in intensities using gradient descent. We implement our algorithm using Pytorch to make use of GPU-acceleration, allowing us to optimize a dataset of slices in under a minute.

Alt text

Citation and Contribution

Please cite this work if any of our code or ideas are helpful for your research.

@inproceedings{TODO,
  title={Intensity-based 3D motion correction for cardiac MR images},
  author={Stolt-Ans{\'o}, Nil and Sideri-Lampretsa, Vasiliki and Dannecker, Maik and Rueckert, Daniel},
  booktitle={TODO},
  year={2024},
  organization={TODO}
}