Room: Exhibit Hall | Forum 5
Purpose: Deformable image registration (DIR) is at the heart of any robust adaptive program in radiation oncology. The purpose of this work is to evaluate SmartAdapt, a DIR platform from Varian, for thoracic cancer patients. Our focus is on the accuracy of CT deformation and contour propagation between two CTs.
Methods: To study CT DIR accuracy, target registration error (TRE) for ten landmarked 4DCT patients from dir-labs (www.dir-lab.com) were calculated. Each of the 10 4DCT images contained 300 landmarks per study matching the inhalation and exhalation phase images, providing a ground truth for the TRE. To study contour propagation accuracy, 20 patients were selected from our own database. Each patient had contours propagated from an original planning CT to a repeat CT using SmartAdapt. The accuracy of the end product was evaluated using the DICE similarity coefficient. The contours evaluated included left and right lung (N=21), heart (N=20), esophagus (N=17) and spinal canal (N=20).
Results: The overall average TRE (including all 10 cases) was 1.3 mm with individual studies having average TREâ€™s ranging between 0.7 and 2.1 mm. The original published results using this data set and the dir-labs own DIR algorithm (PMB 2009) reported a significantly larger average TRE of 1.9 mm.For contour propagation accuracy, the DICE coefficient results for left lung, right lung, heart, esophagus and spinal canal were respectively 0.93, 0.94, 0.90, 0.61, and 0.82 using DIR.
Conclusion: In comparison to the dir-labs DIR from 2009, SmartAdapt shows a superior performance in terms of TRE. SmartAdapt also provides an acceptable level of accuracy for contour propagation considering the DICE coefficients are close to 1. This study demonstrates the validity of SmartAdapt for CT to CT DIR for contour propagation, PET-CT registration and the future prospect of deformed dose accumulation.