Program Information
Assessment of the Accuracy of DIR On MR Images Using Velocity and An In-House Demons Algorithm
R Ger1,2*, J Yang1,2 , Y Ding1 , M Jacobsen1,2 , C Fuller1,2 , R Howell1,2 , H Li1,2 , R Stafford1,2 , S Zhou1,2 , L Court1,2 , (1) The University of Texas MD Anderson Cancer Center, Houston, TX, (2) The University of Texas MD Anderson Cancer Center UTHealth Graduate School of Biomedical Sciences, Houston, TX
Presentations
TU-AB-601-3 (Tuesday, August 1, 2017) 7:30 AM - 9:30 AM Room: 601
Purpose: To evaluate MRI deformable image registration (DIR) error.
Methods: The registration error of Velocity, a commercial b-spline-based software, and an in-house demons-based algorithm was analyzed using (1) a deformable porcine tissue phantom and (2) synthetic images. (1) Porcine phantom: The porcine tissue was implanted with ten 0.35 mm diameter gold markers not visible on MRI. CT, T1-weighted, and T2-weighted images were taken in four different deformed positions. The CT and MRI in each position were rigidly registered and the markers were contoured. The T1-weighted images were deformably registered to the T1-weighted images in the other positions. Similarly, T2-weighted images were deformably registered to the other T2-weighted images. The registration error was measured between the deformably propagated marker contour and the true marker location. (2) Synthetic images: Inter- and intra-patient models were trained on 27 head and neck patients. Four synthetic pre-treatment images were generated using a principal component analysis (PCA)-based inter-patient variation model and their corresponding post-treatment images were generated using a PCA-based intra-patient variation model. The deformation between the pre- and post-treatment images is known. The difference between the known deformation and the deformation generated by the registration system was calculated for evaluation within five regions of interest.
Results: Evaluation of the in-house demons-based algorithm using the porcine phantom images produced root mean square errors (RMSEs) between 1.2-2.1 mm for T1-weighted images and 0.81-1.1 mm for T2-weighted images. Evaluation of Velocity using the porcine phantom images produced RMSEs between 1.5-2.7 mm for T1-weighted images and 1.2-1.6 mm for T2-weighted images. Evaluation of Velocity using the synthetic images produced RMSEs between 0.69-0.76 mm for T1-weighted images and RMSEs between 0.75-1.1 mm for T2-weighted images.
Conclusion: The registration error was less than 3mm for all RMSEs supporting the use of both image registration systems for studies that utilize MRI DIR.
Funding Support, Disclosures, and Conflict of Interest: Rachel Ger is a recipient of the Rosalie B. Hite Graduate Fellowship in Cancer Research awarded by The University of Texas MD Anderson Cancer Center UTHealth Graduate School of Biomedical Sciences.
Contact Email: