3016 - Evaluation of Deformable Image Registration Quality from Contour and Image Intensity Informed Algorithms for Gynecological Cases Undergoing External Beam Radiotherapy and High Dose Rate Brachytherapy Treatments

Purpose/Objective(s): When treating gynecological (GYN) cases with external beam radiotherapy (EBRT) and high dose rate brachytherapy (HDR), registration accuracy is critical for the dose summation of the plan. The purpose of this work is to evaluate a new registration approach which utilizes regional intensity prioritization (REG+) against traditional hybrid contour-intensity based registration (CHB) and intensity-based registration methods (INT).

Materials/Methods: Twenty patients with CT scans from GYN HDR/EBRT treatments were retrospectively selected from a single institution. Bladder and Rectum contours from each plan were used for the contour-based methods. Sigmoid colon was excluded from the analysis due to inconsistent delineation and high degree of observed mobility and deformation. Three deformable image registration (DIR) methods were evaluated: an INT method, a traditional CHB method, and the REG+ method. Each registration method was evaluated using Dice Similarity Coefficient (DSC), Mean Distance to Agreement (MDA), 95th Percentile Hausdorff Distance (HD95); and percentage of structure with Jacobian Determinant less than 0 (JD < 0).

Results: The REG+ method performed best overall for the bladder and rectum, achieving the best contour similarity scores and only 5 out of 20 cases having JD < 0. The INT method had a smaller JD < 0 (0% for bladder and 0% rectum), but the REG+ method had an average of < 1% of the total contour volume having negative values (0.25% for bladder and 0.99% for rectum). It was observed some structures had different segmentation styles between the images which may have contributed to the < 0 JD values. The table provided shows the average (AVG) scores for each method along with the margin of error (ME) for a 95% confidence interval.

Conclusion: Spatial dose summation for GYN HDR/EBRT treatments is currently limited by poor DIR performance. A new method with regional intensity prioritization was compared to traditional approaches and was found to provide overall superior DIR accuracy. Given the metrics are consistent with the acceptable range mentioned in TG-132 (above 0.8 for DSC and less than 2-3mm for MDA), and 15/20 cases had no negative JD values, this method has the potential to further improve registration reliability for GYN HDR/EBRT dose summation. Further investigation could include an expert review of the DIRs, especially in the localizations with JD < 0.