An automatic delineation method of clinical target volume based on prior constraints of organs-at-risk
Objective Deep learning technology has been used for automatic planning of radiotherapy targets,such as directly delineating the patient's clinical target volume(CTV).Current deep learning methods focus on directly predicting the CTV from CT images without considering prior knowledge constraints,which limits the accuracy of such methods.Target delineation is highly related to organs-at-risk(OARs),and existing methods are insufficient in using this prior knowledge.This paper aims to develop an automatic CTV delineation method that integrates prior constraints on organs-at-risk.Methods According to the prior positional relationship between OAR and CTV,the edge distance map is used to enhance the model feature extraction capability.Combined with a multi-task learning framework,a rich set of auxiliary tasks are used to improve the model performance of automatic clinical target delineation.The cervical cancer CTV delineation data of the radiotherapy department of the First Affiliated Hospital of Anhui Province from June 2017 to May 2019 were collected,and the performance of the method was verified on this dataset.Results Experimental results have shown that the proposed automatic CTV delineation method integrating the prior constraints of OAR has achieved good performance on private dataset.The evaluation metrics are better than the existing automatic methods.Conclusion Integrating the prior knowledge of OAR can increase the accuracy of clinical target delineation,which lays the foundation for the automatic target delineation method to enter clinical practice.
万方数据
维普
