Creation of RTOG compliant patient CT-atlases for automated atlas based contouring of local regional breast and high-risk prostate cancers

BACKGROUND Increasing use of IMRT to treat breast and prostate cancers at high risk of regional nodal spread relies on accurate contouring of targets and organs at risk, which is subject to significant inter- and intra-observer variability. This study sought to evaluate the performance of an atlas based deformable registration algorithm to create multi-patient CT based atlases for automated contouring. METHODS Breast and prostate multi-patient CT atlases (n = 50 and 14 respectively) were constructed to be consistent with RTOG consensus contouring guidelines. A commercially available software algorithm was evaluated by comparison of atlas-predicted contours against manual contours using Dice Similarity coefficients. RESULTS High levels of agreement were demonstrated for prediction of OAR contours of lungs, heart, femurs, and minor editing required for the CTV breast/chest wall. CTVs generated for axillary nodes, supraclavicular nodes, prostate, and pelvic nodes demonstrated modest agreement. Small and highly variable structures, such as internal mammary nodes, lumpectomy cavity, rectum, penile bulb, and seminal vesicles had poor agreement. CONCLUSIONS A method to construct and validate performance of CT-based multi-patient atlases for automated atlas based auto-contouring has been demonstrated, and can be adopted for clinical use in planning of local regional breast and high-risk prostate radiotherapy.