Click here to


Are you sure ?

Yes, do it No, cancel

Evaluation of Surface-Based DIR Accuracy Using Digital Phantom Simulations and a Customized Physical Phantom for MR-Guided Plan Adaptation

A Thomas1*, Y Lu2 , P Parikh3 , D Yang1 , (1) Department of Radiation Oncology, Washington University School of Medicine, Saint Louis, MO, (2) Medical College of Wisconsin, Brookfield, WI, (3) Department of Radiation Oncology, Henry Ford Hospital, Detroit, MI


(Sunday, 7/14/2019) 1:00 PM - 2:00 PM

Room: 303

Purpose: Deformable image registration (DIR) is a key prerequisite for a variety of important clinical innovations, in particular those related to advanced radiation therapy (RT). Even with acceptable results for some body regions (e.g. lung), intensity-based DIR methods have shown suboptimal performance when applied to other highly deformable areas such as the GI tract. In this work, we evaluated the accuracy of several surface-based DIR methods such as Shape Context, CPD, and GMM. This enabled an assessment of their potential utility in clinically relevant applications such as multimodality fusion and dose accumulation.

Methods: The digital phantom was created by applying simulation deformations to real patient structure sets via translations, rotations, dilations/contractions, and added noise. The physical phantom was made of two different shapes and sizes of latex synthetic bladders and a variety of different spherical landmarks (63) applied to their surfaces. Point to point correspondences between registered datasets were tracked with selected points from the simulated deformations and the landmarks from the phantom. Mean, target (TRE), and maximum registration errors were compared across the different DIR methods, as well as to more generic registration assessments such as DICE and MHD.

Results: All DIRs tested showed excellent general accuracy for simulated deformations, with certain methods performing better depending on the type of deformation. The registration errors for the physical phantom were much higher, but still comparable to or better than previously reported investigations in general. Mean error and TRE for tracked landmarks were as low as ~3 mm despite raw deformations ranging from 20-70 mm.

Conclusion: A variety of DIR methods’ accuracy was tested with simulated deformations and physical deformations with a phantom. The strengths and weaknesses of the methods tested were identified in an effort to better inform their relevance and feasibility for specific clinical applications.


Not Applicable / None Entered.


Not Applicable / None Entered.

Contact Email