Click here to

Session: Multi-Disciplinary General ePoster Viewing [Return to Session]

Quantitative Analysis of the Accuracy of CT to CBCT Deformable Image Registration for Contour Mapping for Prostate Cancer

R Schmidt*, N Dogan, J Ford, M Studenski, M Abramowitz, K Padgett, B Spieler, Y Xu, R Delgadillo, University of Miami, Miami, FL


PO-GePV-M-118 (Sunday, 7/25/2021)   [Eastern Time (GMT-4)]

Purpose: Deformable Image Registration (DIR) can automatically propagate planning CT (pCT) contours to daily cone-beam CT (CBCT) images to account for anatomical changes such as bladder and rectal filling. The accuracy of deformed contours depends on several factors, including DIR process, algorithm and image modality/quality. This study hypothesizes that auto-propagated prostate contours through the commercially available DIR applied to daily CBCT are sufficiently robust and accurate.

Methods: 28 prostate cancer patients enrolled on an IRB-approved protocol, treated with VMAT and receiving daily CBCT images were selected. Following rigid registration shifts, the pCT was deformably registered to the daily CBCTs using an intensity-based DIR using a B-spline algorithm to generate deformed (auto) prostate contours. Gold standard (reference) contours on CBCT images were manually drawn by an experienced radiation oncologist. DIR algorithm performance was evaluated by comparing the auto and manually-drawn prostate contours on daily CBCT images (1,010 fractions) via dice similarity (DSI), mean distance-to-agreement (MDA), change in center of mass position (ΔCM) and Jacobian. 42 radiomic texture features were extracted from both contours and there correlation was determined through Pearson’s correlation (PCC) and Lin’s concordance correlation (LCC).

Results: The average DSI, MDA and ΔCM between the auto- and manually-drawn contours were found to be 0.90±0.04, 1.81±0.47 mm and 2.17±1.26 mm respectively. The average minimum and maximum of the Jacobian determinant were 0.77±0.18 and 1.30±0.23 respectively. 94.8% of all auto contours were within tolerance provided by TG-132. 34 radiomic texture features were robust to small changes in contours as seen through high PCC and LCC (>0.80), indicating that small contour differences do not significantly change the radiomic texture feature data.

Conclusion: The use of small ROI improved the DIR accuracy for prostate contour propagation to be within tolerance recommendations of TG-132 and appears to be suitable for radiomic analysis and/or dose accumulation.

Funding Support, Disclosures, and Conflict of Interest: This work is supported by Varian Medical Systems. COI - honorarium from Varian.



    Cone-beam CT, Deformation, Registration


    Not Applicable / None Entered.

    Contact Email