Skip to content

Project: MR2CBCT: Restoring and Extending Automated CBCT-MRI Registration for TMJ Analysis #1736

Description

@Dcaleme

Draft Status

Draft - team will hold off on page creation

Category

DICOM

Key Investigators

  • Eduardo Duarte Caleme (University of North Carolina, USA)
  • Lucia Cevidanes (University of North Carolina, USA)
  • Paul Dumont (University of North Carolina, USA)
  • Alex Buisson (University of North Carolina, USA)
  • Steve Pieper (Isomics, USA)
  • Gaelle Leroux (CPE Lyon, France)
  • Juan Prieto (University of North Carolina, USA)
  • Alban Gaydamour (CPE Lyon, France)

Project Description

MRI-to-CBCT registration remains a pressing challenge in medical imaging. Simultaneous visualization of hard and soft tissue structures benefits both clinicians and patients in the diagnosis of temporomandibular degenerative joint disease. Challenges remain in accurately registering these two modalities due to differences in intensity distributions that complicate mutual information optimization, as well as the necessity for initial manual alignment, which can prove unintuitive and challenging for clinicians using current 3D Slicer tools.

Objective

  1. Fix issues with the current registration pipeline that impede it from producing appropriate outputs.
  2. Implement clinician-friendly manual registration tools.
  3. Expand options in both fully automated and clinician-in-the-loop workflows for multi modal registration between MRI and CBCT.

Approach and Plan

  1. Present the restored MR2CBCT pipeline and validated results to the community.
  2. Present, test, and optimize the new clinician-friendly manual registration tools.
  3. Discuss validation framework design.

Progress and Next Steps

  1. Fixed a critical bug in AREG_MRI.py where process_images() was never called.
  2. Fixed naming convention restrictions throughout the pipeline that proved unintuitive for clinicians.
  3. Fixed a mask size mismatch crash in apply_mask.py and a left/right label swap in TMJ crop side detection.
  4. Implemented clinician-friendly manual approximation tools, now implemented as a slicer transforms core feature
  5. Successfully ran the registration pipeline on 25 cases outside of the initial validation sample, yielding clinically satisfying results.
  6. Changed the logic of the approximation step to perform a center of mass calculation in the MRI and a NNUNET segmentation on the CBCT with approximation performed by fiducial registration, a more reliable approach for large distances than torch NMI

Illustrations

No response

Background and References

No response

https://www.youtube.com/watch?v=Qt__yyEzrHk

Image

Initial images

Image

Center MRI on CBCT function

Image

Clinician manual registration

Image

Elastix Refinement

Metadata

Metadata

Assignees

Type

No type

Projects

No projects

Milestone

No milestone

Relationships

None yet

Development

No branches or pull requests

Issue actions