Automatic Segmentation of Clinical Target Volume and Organs-At-Risk for Breast Conservative Radiotherapy Using a Convolutional Neural Network
Abstract Background and Objective: Delineation of the clinical target volume (CTV) and organs at risk (OARs) is very important for radiotherapy but is time-consuming and prone to inter- and intra-observer variation. We trained and evaluated a U-Net-based model to provide fast and consistent auto-segmentation for breast cancer radiotherapy. Methods: We collected 160 patients’ computed tomography (CT) scans with early-stage breast cancer who underwent breast-conserving surgery (BCS) and were treated with radiotherapy in our center. CTV and OARs (contralateral breast, heart, lungs and spinal cord) were delineated manually by two experienced radiation oncologists. The data were used for model training and testing. The dice similarity coefficient (DSC) and 95th Hausdorff distance (95HD) were used to assess the performance of our model. CTV and OARs were randomly selected as ground truth (GT) masks, and artificial intelligence (AI) masks were generated by the proposed model. The contours were randomly distributed to two clinicians to compare CTV score differences. The consistency between two clinicians was tested. We also evaluated time cost for auto-delineation. Results: The mean DSC values of the proposed method were 0.94, 0.95, 0.94, 0.96, 0.96 and 0.93 for breast CTV, contralateral breast, heart, right lung, left lung and spinal cord, respectively. The mean 95HD values were 4.31 mm, 3.59 mm, 4.86 mm, 3.18 mm, 2.79 mm and 4.37 mm for the above structures respectively. The average CTV scores for AI and GT were 2.92 versus 2.89 when evaluated by oncologist A (P=.612), and 2.75 versus 2.83 by oncologist B (P=.213), with no statistically significant differences. The consistency between two clinicians was poor (Kappa=0.282). The times for auto-segmentation of CTV and OARs were 3.88 s and 6.15 s. Conclusions: Our proposed model can improve the speed and accuracy of delineation compared with U-Net, while it performed equally well with the segmentation generated by oncologists.