Room: Exhibit Hall | Forum 9
Purpose: To assess the performance and limitations of a deformable image registration (DIR) algorithm developed on a commercial platform for single-modality, daily CT-on-rails (CTOR) image guided patients as well as multi-modality, daily Cone Beam CT (CBCT) image guided patients.
Methods: 20 prostate cancer patients treated with IMRT/VMAT were selected for analysis. A total of 453 fractions across those patients were analysed. Image data was imported into MIM (v.6.8b, MIM Software, Inc., Cleveland, OH) and the DIR algorithm was applied to register each fraction with the planning CT. The corrected manually drawn contours delineating the bladder and rectum were utilized for comparison against the DIR propagated contours in each fraction. Four metrics were utilized in the evaluation of contour similarity.
Results: 95% Confidence Intervals were assessed via a Linear Mixed Effect model for the Hausdorff Distance, Mean Distance to Agreement, Dice Similarity Coefficient, and Jaccard indices. Hausdorff Distance comparing manual with propagated contours was assessed at 18.47Â±5.87mm for bladder and 17.32Â±2.52mm for rectum. Mean Distance to Agreement was assessed at 4.29Â±1.28mm for bladder and 3.12Â±0.54mm for rectum. Dice Similarity Coefficient was assessed at 0.739Â±0.053 for bladder and 0.736Â±0.032 for rectum. Jaccard was assessed at 0.602Â±0.066 for bladder and 0.593Â±0.04 for rectum. A subfactor analysis was performed per modality (CTOR vs. CBCT) and time (fraction). Statistical significance at Î±=0.05 was observed for manual vs. table shifted deformably propagated contours for rectum for all the metrics except Hausdorff Distance.
Conclusion: The high variance of Hausdorff Distance across all automated methods for bladder indicates widely variable agreement across fractions for all patients. Lower variance across all modalities, methods, and metrics were observed for rectum. Screening of image guidance for substantial differences in bladder and rectal filling compared with the planning CT reference could aid in identifying fractions for which automated DIR would prove insufficient.