Program Information
Comparing Single- and Multi-Atlas Based Segmentation in a Commercial Treatment Planning System
S Andersson*, S Svensson , O Weistrand , RaySearch Laboratories, Stockholm, Sweden
Presentations
WE-RAM3-GePD-I-2 (Wednesday, August 2, 2017) 10:30 AM - 11:00 AM Room: Imaging ePoster Lounge
Purpose: A development version of the atlas based segmentation (ABS) with a multi-atlas approach was compared to the currently available single-atlas approach in the commercial treatment planning system RayStation.
Methods: A pelvic MRI T2-weighted template consisting of data sets from 19 patients was used in the evaluation. This template includes contours of nine structures in the pelvic region. The performance of the ABS using both a single-and a multi-atlas approach was evaluated in a leave-one-out procedure. The segmentation was performed in a development version of RayStation and the computational time for each case was recorded. The Dice Similarity Coefficient (DSC) was used to quantify the performance. In the single-atlas approach, the estimated best atlas is deformed, while in multi-atlas, the N best atlases are deformed and fused together. N can be set by the user and in this evaluation 18 was used. The fusion is done in an iterative process which is initialized with a weighted majority voting.
Results: The computational time was in average 57.5 and 189.7 seconds for the single- and multi-atlas approach, respectively. The mean and median DSC were improved for all structures. The median DSC increased from 0.816 to 0.835, 0.824 to 0.859, and 0.808 to 0.890 for the prostate, rectum, and urinary bladder, respectively.
Conclusion: As the multi-ABS takes only three minutes using a template as large as 18 atlases and with improved performance for all structures, one can argue that it should always be the first step in the delineation process.
Funding Support, Disclosures, and Conflict of Interest: The authors are full-time employees of RaySearch Laboratories.
Contact Email: