Performance of an atlas-based autosegmentation software for delineation of target volumes for radiotherapy of breast and anorectal cancer.

Abstract

To validate atlas-based autosegmentation for contouring breast/anorectal targets. ABAS uses atlases with defined CTVs as template cases to automatically delineate target volumes in other patient CT-datasets. Results are compared with manually contoured CTVs of breast/anorectal cancer according to RTOG-guidelines. The impact of using specific atlases matched to individual patient geometry was evaluated. Results were quantified by analyzing Dice Similarity Coefficient (DSC), logit(DSC) and Percent Overlap (PO). DSC >0.700 and logit(DSC) >0.847 are acceptable. In addition a new algorithm (STAPLE) was evaluated. ABAS produced good results for the CTV of breast/anorectal cancer targets. Delineation of inguinal lymphatic drainage, however, was insufficient. Results for breast CTV were (DSC: 0.86-0.91 ([0,1]), logit(DSC): 1.82-2.36 ([-∞,∞]), PO: 75.5-82.89%) and for anorectal CTVA (DSC: 0.79-0.85, logit(DSC): 1.40-1.77, PO: 68-73.67%). ABAS produced satisfactory results for these clinical target volumes that are defined by more complex tissue interface geometry, thus streamlining and facilitating the radiotherapy workflow which is essential to face increasing demand and limited resources. STAPLE improved contouring outcome. Small target volumes not clearly defined are still to be delineated manually. Based on these results, ABAS has been clinically introduced for precontouring of CTVs/OARs.