基于人工智能与深度学习在脑胶质瘤精准放疗肿瘤靶区及正常器官自动勾画的研究  

Automatic Outlining of Tumor Targets and Normal Organs Based on Artificial Intelligence and Deep Learning in Precision Radiotherapy for Glioma

作　　者：王盛蕾 房中华 王磊 李国庆 泮卫红 Wang Sheng-lei;Fang Zhong-hua;Wang Lei;Li Guo-qing;Pan Wei-hong(Cyberknife Radiotherapy Center,Zibo Shibo High-tech Hospital,Zibo 255035,Shandong Province,China;CT Room,Zibo First Hospital,Zibo 255200,Shandong Province,China;Information Department,Zibo First Hospital,Zibo 255200,Shandong Province,China;Department of Oncology,Zibo Bashan Wanjie Hospital,Zibo 255200,Shandong Province,China)

机构地区：[1]淄博世博高新医院射波刀放疗中心,山东淄博255035 [2]淄博市第一医院CT室,山东淄博255200 [3]淄博市第一医院信息科,山东淄博255200 [4]淄博岜山万杰医院肿瘤科,山东淄博255200

出　　处：《中外医药研究》2022年第11期66-68,119,共4页JOURNAL OF CHINESE AND FOREIGN MEDICINE AND PHARMACY RESEARCH

摘　　要：目的:人工智能与深度学习在脑胶质瘤精准放疗肿瘤靶区及正常器官自动勾画的研究。方法:选取2019年1月-2021年6月淄博世博高新医院收治的行脑胶质瘤术后放射治疗的患者48例为研究组,将48例脑胶质瘤患者数据分别导入50例和80例数据库中进行靶区及危及器官自动勾画,比较影像诊断医师与放疗医师勾画在数据库数量的增多情况下,肿瘤区(GTV)和肿瘤临床靶区(CTV)肿瘤靶区及正常器官勾画的精度及重复性是否产生影响,比较模型相似度、准确度,勾画效率、勾画误差,对此模型库数据进行理论验证和模型修正。结果:影像医师勾画GTV边缘相似度较放疗医师CTV边缘相似度好,80例数据较50例数据更稳定,均>90%。正常器官分割效果较为稳定,GTV和CTV比较,差异有统计学意义(P<0.05)。结论:两个模型的预测效能和不同表现的数据显示,最终确定,更准确、更高效地提高勾画效率、减少勾画误差的潜在优势。有助于减少照射体积,提高靶区准确性并对放射性脑坏死发生率评定,可为脑胶质瘤放疗的靶区勾画提供临床研究基础。Objective:The study of the automatic outlining of tumor target area and normal organs by artificial intelligence and deep learning in precision radiotherapy for glioma.Methods:48 patients admitted to Zibo Shibo High-tech Hospital who underwent postoperative radiation therapy for glioma from January 2019 to June 2021 were selected as the study group.The data of 48 patients with glioma were imported into the 50 case and 80 case databases for automatic outlining of target areas and organs at risk,respectively.To compare the accuracy and reproducibility of gross target volume(GTV)and tumor clinical target volume(CTV)tumor target area and normal organ outlining by diagnostic imaging physicians and radiotherapists with the increase in the number of databases.We compared the model similarity and accuracy,outline efficiency and outline error,and performed theoretical validation and model correction on this model library data.Results:The similarity of GTV edge outlined by imaging physicians was better than that of CTV edge outlined by radiotherapists,and the data of 80 cases were more stable than that of 50 cases,all of which were>90%.The effect of normal organ segmentation was more stable,and the difference was statistically significant when comparing GTV and CTV(P<0.05).Conclusion:The predictive efficacy of the two models and the different performance data showed the potential advantages of a more accurate and efficient outlining efficiency and reduction of outlining errors,as finally determined.It helps to reduce irradiation volume,improve target accuracy and rate the incidence of radiation brain necrosis,and may provide a basis for clinical studies on target outlining for glioma radiotherapy.

关 键 词：人工智能 深度学习 脑胶质瘤精准放疗 自动勾画 

分 类 号：R73[医药卫生—肿瘤]