Related Experiment Video
Updated: Jun 18, 2026
07:13
Multimodal Cross-Device and Marker-Free Co-Registration of Preclinical Imaging Modalities
Published on: October 27, 2023
Results of a multi-institution deformable registration accuracy study (MIDRAS)
1Princess Margaret Hospital, University Health Network, Department of Radiation Oncology, University of Toronto, Toronto, Ontario, Canada. kristy.brock@rmp.uhn.on.ca
International Journal of Radiation Oncology, Biology, Physics
|November 14, 2009
Summary
Multi-institutional deformable image registration algorithms showed varied accuracy, with most achieving voxel-level precision. This advancement holds promise for improving radiation treatment planning and delivery.
Area of Science:
- Medical imaging
- Computational anatomy
- Radiotherapy research
Background:
- Deformable image registration (DIR) is crucial for aligning medical images.
- Accurate DIR is essential for effective radiation therapy planning and delivery.
- Standardized assessment of DIR algorithms across institutions is lacking.
Purpose of the Study:
- To evaluate the accuracy, reproducibility, and computational performance of DIR algorithms.
- To compare multiple DIR algorithms developed at different institutions using common datasets.
- To establish an infrastructure for assessing multi-institution DIR accuracy.
Main Methods:
- Utilized datasets from lung (4D-CT), liver (4D-CT, MRI), and prostate (MRI) patients.
- Radiation oncologists annotated anatomical structures for ground truth.
- Algorithm accuracy was measured by comparing predicted displacements to expert annotations.
- Invited 37 institutions and manufacturers with published DIR capabilities.
Main Results:
- 27 groups participated; 6 did not return results.
- Completion rates varied: 16 for liver 4D-CT, 12 for lung 4D-CT, 3 for prostate MRI, 3 for liver MRI-CT.
- Reported average absolute errors ranged from 0.4 mm to 6.2 mm across different datasets and directions (LR, AP, SI).
Conclusions:
- Developed a framework for assessing multi-institution DIR accuracy.
- Observed significant discrepancies in reported registration shifts.
- Most DIR algorithms performed at accuracy levels comparable to voxel size, indicating potential for improved clinical applications.
