Program Information
Geometric Correlation Analysis and Dosimetric Plan Quality Evaluation of Contours Produced by An Automatic Atlas-Based Segmentation Tool
B Stiehl1*, M Cao2 , D Ruan3 , (1) ,,,(2) UCLA School of Medicine, Los Angeles, CA, (3) UCLA School of Medicine, Los Angeles, CA
Presentations
SU-F-FS4-3 (Sunday, July 30, 2017) 2:05 PM - 3:00 PM Room: Four Seasons 4
Purpose: To investigate whether atlas-based auto-segmentation (AS) is capable of producing contours that perform comparably to manually segmented (MS) contours in terms of both geometric agreement and dosimetric plan quality endpoints.
Methods: A 48-subject prostate atlas was created and another 15 prostate patient CT image sets were used for testing. Contours were generated using MIM’s atlas-based segmentation, and compared to their clinical counterparts. Geometric correlation was evaluated using Dice similarity coefficient (DSC) and Hausdorff distance (HD). Dosimetric relevance was evaluated by assessing the DVH differences derived by imposing AS and MS on the same planning dose distribution, respectively. Paired t-test was employed. Difference in plan acceptance based on clinical dosimetric endpoints was evaluated for specificity and sensitivity. The comparison was repeated for head/neck (HN) with a 31-subject atlas and 15 testing cases.
Results: Geometric agreement between AS and MS differed significantly across structures: from (L: 0.92 /R: 0.91) for the femoral heads and 0.98 for the brain, to average values of 0.38, 0.36, and 0.42 for the seminal vesicle, chiasm, and pharynx, respectively. Despite the geometric disagreement, the paired t-tests showed insufficient evidence for statistically significant difference (p>0.05) in dosimetric plan quality values yielded by the AS and MS approach for all structures except prostate CTV (p = 0.0096). In HN cases, statistically significant differences in dosimetric end points were observed in only four of the eighteen measurements including cord maximum dose (p = 0.00089), parotid mean dose (p= 0.0222), pharynx mean dose (p = 0.0097), and pharynx V45 (p = 0.032).
Conclusion: Variations of AS performance among structures allude to a differential approach of using AS on a structure subset and focus MS on the rest. The discrepancy between geometric and dosimetric-end-point driven evaluation also indicates the potential utility of AS contours in predicting plan quality notwithstanding geometric imprecision.
Funding Support, Disclosures, and Conflict of Interest: 2017 AAPM Summer Undergraduate Fellowship Program
Contact Email: