Encrypted login | home

Program Information

Validating Deformable Image Registration as a Clinical Tool for Dose Tracking and Adaptive Planning

A Soldner

A Soldner1*, J McGlade1 , H Chen1 , Z Yu2 , F Mourtada1 , (1) Christiana Care Hospital, Newark, DE, (2) Baylor College of Medicine, Houston, TX


SU-I-GPD-J-67 (Sunday, July 30, 2017) 3:00 PM - 6:00 PM Room: Exhibit Hall

Purpose: To validate the accuracy of automated deformable image registration using a novel commercial phantom. Deformable registration of clinical CBCT images can then be used for dose tracking and adaptive planning.

Methods: A CIRS Prostate Training Phantom (Model 070S) was used to determine the accuracy of the RayStation 5.0 deformable registration by tracking interfraction anatomical changes when a random deformation was applied. The CIRS phantom has deformable tissue structures (prostate, seminal vesicles (SV), bladder, and rectum) and was imaged using both simulation CT and CBCT (3-mm thickness). Deformable registration was performed using RayStation’s Hybrid Intensity and Structure Based method, which minimizes an objective function using image similarity terms and anatomical structures. The prostate+SV was contoured manually on each image for comparison with the targets mapped using deformable registration. The overall volume and region of overlap were recorded.

Results: We observed variation in the quality of deformable registration that depended on the anatomical information provided with each image set. Controlling ROIs improve the accuracy of the volume of overlap by 5.29% for the CT and 5.7% for CBCT images. Volume differences for the prostate+SV are under 3.0% for both CT and CBCT, but not necessarily aligned. Using the deformable registration without controlling ROIs can result in errors up to 10% of missing volume overlap.

Conclusion: Current deformation algorithms can produce accurate deformed ROIs, but are highly dependent on image quality and deformation technique. We developed a simple method, using a brachytherapy prostate phantom, to validate RayStation’s deformable image registration. Controlling ROIs are needed to reduce error to within 5%. Moving forward, we plan to use RayStation’s Model Based Segmentation to automatically contour bladder and rectum as control ROIs on CBCTs and tests the deformation algorithm on retrospective patient images. The ability to accurately map deformed ROIs is critical for adaptive planning.

Contact Email: