Program Information
Evaluation of the Accuracy of Deformable Image Registration of Individual Structures Between CT and CBCT in Prostate Cancer Radiotherapy
J Hammers1*, S Pirozzi2 , J Matney1 , O Kaidar-Person1 , R Chen1 , A Wang1 , S Das1 , P Mavroidis1 , (1) University of North Carolina, Chapel Hill, NC, (2) MIM Software Inc., Cleveland, OH
Presentations
WE-RAM2-GePD-JT-6 (Wednesday, August 2, 2017) 10:00 AM - 10:30 AM Room: Joint Imaging-Therapy ePoster Theater
Purpose: To evaluate the segmentation contours generated using an automated deformable registration algorithm based on the planning CT and daily CBCTs against those obtained through the manual segmentation of those contours on each CBCT by a radiation oncologist.
Methods: Twenty two CBCTs were selected from a single prostate cancer patient. The contours of CTV, bladder, and rectum were manually defined by a radiation oncologist on each CBCT, and were used to apply a masking technique to highlight the pixels corresponding to bladder and rectum in each CBCT as well as the planning CT in order to enhance the registration in these organs. A normalized deformable registration method was applied to deform the contours from the planning CT to each CBCT for both the original data and the bladder-rectum highlighted (BRH) data. The average Dice Similarity Coefficients (DSC) and mean and max Hausdorff distances were calculated between the deformed planning CT contours (dPC) and the BRH deformed planning CT contours (BRH–dPC) for each structure to be compared against the manually defined “gold” standard contours for each CBCT. The overall percent improvement was calculated as the proportion of the error corrected by the method.
Results: The DSC values showed significantly superior agreement in the BRH process between the deformed and manual contours compared to the original data for bladder and rectum (p<0.05). The mean DSCs for the bladder and rectum original data were 0.85±0.07 and 0.75±0.07, respectively, while the values for the BRH-processed data were 0.94±0.04 and 0.90±0.04, respectively. The mean DSCs for the CTV were not significantly different between the two methods (p>0.05).
Conclusion: The high-contrast boundary generated in the BRH-process method resulted in significantly improved deformable registration in bladder and rectum. The agreement between the deformed and physician’s contours increased by approximately 60% for bladder and 62% for rectum.
Contact Email: