<p>Commercial software tools for automatic segmentation have been adopted in breast cancer radiotherapy planning. In this study, we directly compared commercial software programs for single-atlas-based segmentation (SABS), multi-atlas-based segmentation (MABS), and deep-learning-based segmentation (DLBS) on the ipsilateral breast using Eclipse v16, RayStation 16, and syngo.via RT Image Suite VB80G, respectively. Thirty previously registered atlases were stratified into three equal groups based on the mean bilateral breast height. An additional 20 patients with breast cancer who underwent radiotherapy after breast-conserving surgery were included retrospectively. For MABS, a parameter-matched tertile dataset was selected for each of the 20 test cases. For SABS, three of the 30 atlases were registered, and one of the registered atlases was selected for each test case by considering both nipple position and breast height. Manual segmentation was independently performed by two radiation oncologists. The Dice similarity coefficient (DSC) was used to assess agreement between two observers and between manual and automatic segmentations. The Friedman test was used to compare the DSC values of the automatic methods. The median DSC for interobserver agreement was 0.92. Although DSC values obtained using SABS were significantly lower than those of the other methods, its computation time was the shortest among the automatic methods. Despite the requirement for atlas datasets, MABS exhibited DSC values comparable to those of DLBS, with median DSC values of approximately 0.9 obtained in both methods. Clinical implementation can be optimized according to individual institutional requirements by understanding the strengths and limitations of each method.</p>