Evaluation of Atlas Based Segmentation in Accelerated Partial Breast Irradiated Patients.

Abstract

PURPOSE/OBJECTIVE(S)\nAccelerated partial breast irradiation (APBI) provides a convenient treatment option for patients with early-stage breast cancer, reducing treatment duration and the volume of breast treated. Automated contouring tools can decrease contouring time and improve consistency in contours of organs at risk (OARs). The purpose of this study was to evaluate the efficacy of atlas-based contours created by a commercially available software compared with manually contoured left breast, right breast, and heart OARs.\n\n\nMATERIALS/METHODS\nThirty-one patients treated with APBI were randomly selected to generate an OAR atlas. The manually contoured left breast, right breast, heart, left lung, right lung, and spinal cord were included to construct the atlas using commercially available software. Since the atlas-based contour for lungs and spinal cord are well established at our clinic, our evaluation focused on the left/right breast and heart OARs. Using a specific workflow for partial breast irradiation, atlas-based OAR contours were automatically generated upon uploading of the image set. For left and right breast contours, this workflow automatically contracted breast contours 5mm away from skin surface. A twenty-one patient dataset was used to test the atlas. Dice Similarity Coefficient (DSC) and the mean Hausdorff distance (MSHD) were used for quantitative comparison to contoured breast and heart OARs. In addition to quantitative analysis, atlas-based segmentation was evaluated qualitatively by two staff radiation oncologists. They evaluated the contours using a four scaled score, defined as no change, minor change, major change, or not acceptable.\n\n\nRESULTS\nThe median and range DSC for the left breast, right breast and the heart were 0.85 ± 0.06 (0.71-0.94), 0.85 ± 0.06 (0.77-0.94), and 0.93 ± 0.03 (0.83-0.97), respectively. The average MSHD were 0.39 cm for the left breast, 0.36 cm for the right breast, and 0.20 cm for the heart. Qualitatively, among 21 test cases, each contour was assigned to whichever was the lower score from the two radiation oncologists. No atlas-based contours were rated not clinically acceptable by the radiation oncologists, consistent with all DSC > 0.7. 71% of atlas-based left breast contours were rated minor change, 81% for the right breast, and 76% for the heart, also consistent with the DSC values observed above.\n\n\nCONCLUSION\nAtlas-based segmentation used for patients receiving APBI is clinically useful with minor contour editing required from physicians. Our results from using a commercially available software program can be readily applied to other centers and may be tested for patients undergoing whole breast irradiation.