Study on Dose Distribution Prediction of Esophageal Cancer Patients Using U-Net Model

摘要

Accurate radiotherapy for esophageal cancer is very important for medical diagnosis and human health. In the course of radiotherapy, the distribution of dose in different organs is a problem that doctors and patients are very concerned about. In this study, the U-Net model was used to predict the desired dose distribution from the patient's contours of planning target volume (PTV) and organs at risk (OAR), so as to realize the automatic prediction of dose distribution. The object of this study was the CT data of esophageal cancer. The contours of patients' organs (PTV, left lung, right lung, heart) and dose distribution maps were extracted and processed, then we used U-net model to train the data. For a patient, the network model was trained on a single slice, and the output was the prediction of the dose for the patient slice. By training the network model, the radiotherapy plan was automatically formulated based on prior knowledge. Cosine similarity was used to evaluate the similarity between the predicted dose distribution maps and the true dose distribution maps. In addition, the dose volume histogram (DVH) curve of the true dose and the predicted dose was used to evaluate the accuracy of the prediction. It can be seen that the dose distribution predicted by U-Net is very close to the true dose distribution.