Dose Prediction for Cervical Cancer Brachytherapy Using 3-D Deep Convolutional Neural Network
Ming Ma, Elizabeth Kidd, B Fahimian, Bin Han, Thomas Niedermayr, Dimitre Hristov, Lei Xing, Yong Yang
Abstract
Brachytherapy is a critical modality in treatment of cervix cancer. However, developing a clinically acceptable plan is a time-consuming manual process. In this article, we propose a dose prediction method for cervix cancer brachytherapy using a 3-D deep convolutional neural network (CNN) to guide treatment planning. The method learns the prediction model from a database of historical treatment plans and predicts the 3-D dose distribution with the 3-D organ structures and positions of the inserted applicators as inputs. 110 treatment plans for cervical cancer brachytherapy were used for training the deep CNN to learn a 3-D dose prediction model. Twenty cases were used to evaluate the performance of the proposed method by analyzing the outcomes in terms of mean dose values, derived DVHs from dose distribution, and statistical analysis of sum of absolute residual. Our results show that the predicted mean V100 of the CTV and mean D2cc of organ-at-risks are close to those in the clinical treatment plans. The mean D2cc doses in bladder, rectum and bowel are 5.97, 3.67, and 4.05 Gy, respectively, in comparison to 5.77, 3.42, and 4.16 Gy in clinical plans, indicating a great potential to predict accurate dose distribution and facilitate the brachytherapy treatment planning workflow.