Related Experiment Video
Updated: Nov 12, 2025
Author Spotlight: An Efficient and Robust Software for Automated Fusion of Multiple Preclinical Imaging Modalities
Published on: October 27, 2023
Benchmarking of Deformable Image Registration for Multiple Anatomic Sites Using Digital Data Sets With Ground-Truth
Liting Shi1, Quan Chen2, Susan Barley3
1Department of Radiation Oncology, University of California Davis Medical Center, Sacramento, California; School of Biomedical Engineering (Suzhou), Division of Life Sciences and Medicine, University of Science and Technology of China, Hefei, Anhui, China; Department of Medical Imaging, Suzhou Institute of Biomedical Engineering and Technology, Chinese Academy of Science, Suzhou, Jiangsu, China.
Deformable image registration (DIR) accuracy varies by anatomical site and deformation intensity. Contour-based metrics often do not correlate with deformation vector field (DVF) errors, highlighting the need for site-specific tolerance values in adaptive radiation therapy.
Area of Science:
- Medical physics
- Radiotherapy technology
- Image analysis
Background:
- Deformable image registration (DIR) is crucial for adaptive radiation therapy (ART).
- Accurate assessment of DIR performance requires understanding the relationship between deformation vector field (DVF) errors and contour-based metrics.
- Existing studies often lack comprehensive evaluation across diverse anatomical sites and deformation levels.
Purpose of the Study:
- To evaluate the accuracy of DIR algorithms using datasets with varying ground-truth deformation vector fields (DVFs).
- To investigate the correlation between DVF errors and contour-based metrics in different anatomical regions.
- To provide benchmark data for assessing DIR accuracy in various clinical scenarios.
Main Methods:
- Generated nine digital datasets with controlled deformations (low, medium, high intensity) from anonymized patient CTs (head/neck, thorax/abdomen, pelvis) using ImSimQA software.
- Performed DIR and contour propagation using MIM-Maestro, Raystation, and Velocity systems.
- Compared system-generated DVFs and propagated contours against ground-truth data; evaluated correlations using Pearson (r) and Spearman (rho) coefficients.
Main Results:
- DVF errors increased with deformation intensity.
- DIR algorithms performed well for esophagus, trachea, and pelvic structures (mean DVF errors <2.50 mm).
- Significant DVF errors were observed for brain, liver, lung, and bladder (dmax: 2.8-91.90 mm).
- Contour metrics correlated with volumes, particularly for small/large structures, but showed limited correlation with DVF errors.
- Only Raystation and Velocity showed moderate correlation between distance metrics and DVF errors (r: 0.70-0.78).
Conclusions:
- Most contour-based metrics do not accurately reflect DVF errors.
- Accurate contour propagation does not guarantee accurate dose summation in ART.
- Acceptable DVF error tolerances should be site-specific, considering deformation intensity and organ size.
- This study offers benchmark tables for evaluating DIR accuracy in diverse clinical contexts.
