Abstract
Purpose:To investigate the effectiveness of atlas selection methods for improving atlas‐based auto‐contouring in radiotherapy planning.Methods:275 H&N clinically delineated cases were employed as an atlas database from which atlases would be selected. A further 40 previously contoured cases were used as test patients against which atlas selection could be performed and evaluated. 26 variations of selection methods proposed in the literature and used in commercial systems were investigated. Atlas selection methods comprised either global or local image similarity measures, computed after rigid or deformable registration, combined with direct atlas search or with an intermediate template image. Workflow Box (Mirada‐Medical, Oxford, UK) was used for all auto‐contouring. Results on brain, brainstem, parotids and spinal cord were compared to random selection, a fixed set of 10 “good” atlases, and optimal selection by an “oracle” with knowledge of the ground truth. The Dice score and the average ranking with respect to the “oracle” were employed to assess the performance of the top 10 atlases selected by each method.Results:The fixed set of “good” atlases outperformed all of the atlas‐patient image similarity‐based selection methods (mean Dice 0.715 c.f. 0.603 to 0.677). In general, methods based on exhaustive comparison of local similarity measures showed better average Dice scores (0.658 to 0.677) compared to the use of either template image (0.655 to 0.672) or global similarity measures (0.603 to 0.666). The performance of image‐based selection methods was found to be only slightly better than a random (0.645). Dice scores given relate to the left parotid, but similar results patterns were observed for all organs.Conclusion:Intuitively, atlas selection based on the patient CT is expected to improve auto‐contouring performance. However, it was found that published approaches performed marginally better than random and use of a fixed set of representative atlases showed favourable performance.This research was funded via InnovateUK Grant 600277 as part of Eurostars Grant E!9297.DP,BS,MG,TK are employees of Mirada Medical Ltd.
Sign-in/Register to access full text options 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a callDisclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.
