(Wednesday, 8/1/2018) 4:30 PM - 6:00 PM
Room: Karl Dean Ballroom C
Purpose: To address the challenges faced by single-cycle respiratory-correlated 4DCT, a variety of motion models have been proposed. These models aim to capture long-term, cycle-to-cycle variations in the thoracic anatomy, and accurately characterize the impact of complex motions associated with the tumor and organs in the irradiated volume. A critical requirement for the clinical translation of such models is to put in place a reliable patient- and motion-specific validation-framework to test the intra- and interfractional accuracy and reproducibility of the model. In this work, we develop such a validation framework using kV fluoroscopy, and demonstrate its use on a surface photogrammetry+CT motion model developed for a lung cancer patient.
Methods: Under IRB-approval, a lung SBRT patient was CT scanned while a prototype visionRT system monitored the thoracoabdominal surface. Diffeomorphic image registration was used to relate the exhale phase of the 4DCT to the remaining phases. A principal least square model was trained with the surfaces and the 4DCT deformation vector fields (DVF). During the patient's fourth treatment, fluoroscopic images were performed concurrent with surface monitoring. For each captured surface, the model generated a volumetric CT. A raytracing algorithm was used to create a digitally reconstructed radiograph (DRR) which was compared with fluoroscopic measurements. Displacement of three landmarks (GTV, liver and stomach) contoured in fluoroscopic images by a radiation oncologist was tracked and compared with the model as a quantitative validation.
Results: Liver motion range defined as difference in extrema displacements was: 35mm/11mm(SI/AP) from fluorscopy and 32mm/7mm predicted by model. GTV SI/AP motion range was:8mm/9.6mm by fluoroscopy while 11mm/4mm by the model. RMSE errors for liver centroid was 3.0mm/3.5mm for SI/AP displacement estimation while RME for GTV was 5mm.
Conclusion: We have created a model-agnostic validation-framework that can quantitatively characterize the day-to-day accuracy and robustness our model and other independently-developed models.
Funding Support, Disclosures, and Conflict of Interest: This work was supported by NIH R01CA169102, Varian Medical Systems, Vision RT Ltd.
Keywords
Image-guided Therapy, Modeling
Taxonomy
Not Applicable / None Entered.
Contact Email
