Program Information
Dosimetric Indices From Automated Planning Versus Geometric Conformity Indices for the Evaluation of Contour Accuracy in Head-And-Neck Cancer
T Lim*, E Gillespie , J Murphy , K Moore , UC San Diego, La Jolla, CA
Presentations
SU-I-GPD-T-368 (Sunday, July 30, 2017) 3:00 PM - 6:00 PM Room: Exhibit Hall
Purpose: When quantifying contouring accuracy, geometric indices of conformity to gold-standard contours are commonly used. While convenient, these metrics lack clinical context in radiotherapy. The purpose of this study is to evaluate the correlation between commonly-used geometric indices of contouring accuracy and dosimetric endpoints obtained by generating automated plans using test contours.
Methods: One CT image-set of a nasopharyngeal carcinoma (NPC) patient was used. A gold-standard structure set containing clinical target volumes (CTVs; high-, intermediate- and low-dose), bilateral parotids, bilateral cochleae and other organs-at-risk was defined. Contour sets containing CTVs, parotids and cochleae were obtained from 22 senior physician-residents. Planning target volumes (PTVᴴᴵᴳᴴ;PTVᴵᴺᵀ;PTVᴸᴼᵂ) were generated with standard 3mm margins. 23 treatment plans (22 test, 1 gold-standard) were generated using automated knowledge-based planning to standardize planning output. Critically, while plan optimization was based on test contours, dosimetric indices (PTVs: dose to the hottest 98% volume, D98%; parotids/cochleae: mean dose) were calculated on gold-standard contours. Commonly-used geometric indices (Dice coefficients, Hausdorff max/mean/median distances, volume differences and centroid distances) were calculated. Correlation between dosimetric and geometric indices was quantified with the R² statistic.
Results: For PTVᴴᴵᴳᴴ, volume difference between the gold-standard and test contours had the highest correlation with D98% (R²=0.65). For PTVᴵᴺᵀ, all geometric indices scored R²<0.4 against D98%. For PTVᴸᴼᵂ, mean/median Hausdorff distances were strongly-correlated with D98% (R²>0.7). Compared to ipsilateral parotid (max R²=0.17; Hausdorff median), contralateral parotid showed higher correlation (max R²=0.47; Dice coefficient) for mean dose. For both cochleae, none of the geometric indices reasonably explain the variation in mean doses (R²<0.19 for all).
Conclusion: For this test case in NPC, no one geometric index best explained the clinical consequences of contour variations, and most were weakly-correlated with critical dosimetric endpoints. When available, dosimetric indices generated from standardized automated planning can replace geometric conformity calculations for radiotherapy contouring studies.
Funding Support, Disclosures, and Conflict of Interest: EG and JM acknowledge funding support from AHRQ related to this work. KLM acknowledges research grants, travel funding, and honoraria from Varian Medical Systems.
Contact Email: