北京生物医学工程

基于深度学习方法的放疗患者摆位误差预测

Prediction of setup errors for patients treated with radiotherapy based on deep learning method

作者：高翔宋双张伟陈妙然夏宇曹征

单位：合肥市第一人民医院血液肿瘤科(合肥 230001)<p>通信作者:曹征,高级工程师 E-mail:caozheng81@ 126.com</p>

关键词：深度学习; 放疗; 摆位误差; 锥形束 CT; 预测

分类号：R318.6; R815.6

出版年·卷·期（页码）：2020·39·2（380-388）

摘要：

目的为了实现对放疗患者日常摆位误差的准确预测,优化锥形束计算机断层扫描( cone beam computed tomography,CBCT)使用频率,确保患者摆位误差处在允许范围内的同时尽可能减少其承受的额外辐射剂量,本研究基于深度学习方法能够对复杂体系进行预测的能力,构建了一种深度全连接神经网络? 方法选取 20 名头颈部肿瘤患者累积共 76 次 CBCT 扫描结果作为研究对象? 首先,根据文献调研及临床实践经验,确定患者日常治疗时的摆位误差受到不同技术人员的工作经验?患者体型?固定膜的松紧程度?靶区位置?靶区大小及形状等因素的影响,且与患者前三次治疗时的摆位误差相关性较强,因此针对患者每次 CBCT 扫描得到的摆位误差,从患者的治疗记录及患者 CT 图像中获得相关信息,得到 467 个特征值作为深度学习的输入值,以三个方向上最大摆位误差的分类作为深度学习的输出值,将每次摆位误差的最大值以 3 mm 为标准分为两类,作为深度学习的目标值? 然后,将研究数据按7 ∶ 3 的比例随机划分为训练集和验证集,通过训练集训练深度神经网络,再使用验证集对神经网络进行初步评估? 在完成深度神经网络的训练后,将正在治疗中的新患者的数据作为测试集,使用训练好的神经网络预测新患者的摆位误差大小,并与实际结果对比,评估其准确率? 最后,进行重复实验,判断神经网络的预测结果是否具有可重复性? 结果本研究构建的深度神经网络对患者摆位误差的预测准确率可以达到 86%,能准确预测患者摆位误差大于 3 mm 的情况,且预测结果的可重复性好? 结论基于深度学习方法可以较准确地预测放疗患者日常摆位误差的最大绝对值是否大于 3 mm,为优化 CBCT 扫描频率提供了可靠依据,有助于提高放疗疗效,减轻放疗副反应,具有良好的临床应用价值?

Objective By accurately predicting the daily setup errors of each radiotherapy patient, the use of cone beam computed tomography (CBCT) could be optimized to reduce the risks of additional radiation exposure for the patients. For this purpose, a deep fully connected neural network was proposed in this paper. Methods A total of 76 CBCT results from different patients with head and neck tumors were selected for this study and. First, based on literature research and clinical practice experience, it was determined that the setup errors of patients during daily treatment were affected by the work experience of different technicians, the patients’ body shape, the tightness of the fixed membranes, the locations, the sizes and shapes of the tumors. And it had a strong correlation with the setup errors during the first three treatments of the patients. Therefore, for each CBCT verification result of the patients, relevant information was obtained from the patients’ treatment records and the patients’ CT images, and 467 characteristic values were obtained. A set of 467 feature values was used as the input values of deep learning, and the maximum values of the setup errors in three directions were used as the output values. The maximum value of each setup error was divided into two categories with the standard of 3 mm as the label for deep learning. Then, the research data was randomly divided into a training set and a validation set according to the ratio of 7:3. The deep neural network was trained by the training set, and evaluated by the validation set. After completing the training of the neural network, the data of new patients were used as the test set. The trained neural network and test set were used to predict the daily setup errors of new patients, and the results were compared with the actual results to evaluate its prediction accuracy. Finally, repeated experiments were performed to determine whether the training and prediction results of the neural network were repeatable. Results The prediction accuracy of the patient’s setup errors in this study could reach 86%. The situation that the patients’ setup errors would be greater than 3 mm could be accurately predicted. The test results were reproducible. Conclusions Based on deep learning method, whether the maximum absolute value of the daily setup errors of radiotherapy patients would be greater than 3mm could be accurately predicted, and the accuracy could reach 86%.This study could help departments optimize the frequency of the using of CBCT, improve the efficacy, reduce the side effects of radiotherapy and have good clinical application value.

参考文献：

[ 1 ] 戴建荣, 胡逸民. 图像引导放疗的实现方式[J]. 中华放射肿瘤学杂志, 2006, 15(2):132-135.

Dai JR, Hu YM. Image guided radiotherapy implementation[ J]. Chinese Journal of Radiation Oncology, 2006, 15(2):132-135.

[ 2 ] Kim J, Fessler JA, Lam KL, et al. A feasibility study of mutual information based setup error estimation for radiotherapy [ J ]. Medical Physics, 2001, 28(12):2507-2517.

[ 3 ] Sykes JR, Lindsay R, Iball G, et al. Dosimetry of CBCT: methods, doses and clinical consequences [ J ]. Journal of Physics:Conference Series, 2013, 444:012017.

[ 4 ] Altunbas C, Zheng D, Weiss E, et al. SU - FF - T - 490: the effect of daily cone beam CT imaging dose on the secondary cancer risk for patients receiving prostate IMRT treatments[ J]. Medical Physics, 2009, 36(6Part16):2636.

[ 5 ] Spezi E, Downes P, Jarvis R, et al. Patient-specific three-dimensional concomitant dose from cone beam computed tomography exposure in image-guided radiotherapy [ J ]. International Journal of Radiation Oncology? Biology? Physics, 2012, 83(1):419-426.

[ 6 ] 吴润叶, 高黎, 李明辉, 等. 应用非每日锥形束 CT 校位减少摆位误差对鼻咽癌调强放疗剂量影响[ J]. 中华放射肿瘤学杂志, 2011, 20(5):379-383.

Wu RY, Gao L, Li MH, et al. Using non-daily cone-beam computed tomography reduce dosimetric effect of set-up errors in intensity-modulated radiotherapy for nasopharyngeal cancer [ J]. Chinese Journal of Radiation Oncology, 2011, 20(5):379-383.

[ 7 ] 孔玲玲, 程健, 李宝生, 等. 保留乳房术后放疗锥形束 CT 引导系统摆位误差预测的可行性研究[J]. 中华肿瘤防治杂志,2013, 20(12):949-952.

Kong LL, Cheng J, Li BS, et al. Feasibility study of systematic setup error prediction for breast irradiation after breast-conserving therapy based on cone-beam CT setup verification[ J]. Chinese Journal of Cancer Prevention and Treatment, 2013, 20 ( 12): 949-952.

[ 8 ] 王玮, 李建彬, 徐敏, 等. 锥形束 CT 引导全乳调强放疗摆位误差自适应的预测与校正[ J]. 中华肿瘤杂志, 2016, 38 (3):197-201.

Wang W, Li JB, Xu M, et al. Cone beam CT-derived adaptive radiotherapy for setup error assessment and correction in whole breast intensity modulated radiotherapy [ J]. Chinese Journal of Oncology, 2016, 38(3):197-201.

[ 9 ] 李庆, 尹龙斌, 谢慧轻, 等. 鼻咽癌调强放疗中摆位误差的变化趋势:基于千伏级锥形束 CT 的前瞻性研究[ J]. 放射学实践, 2017, 32(8):870-875.

Li Q, Yin LB, Xie HQ, et al. Trends of inter-fractional setup errors in intensity-modulated radiotherapy for nasopharyngeal carcinoma: a prospective study based on kilovoltage cone-beam computed tomography[ J]. Radiologic Practice, 2017, 32( 8): 870-875.

[10] 丘敏敏, 钟嘉健, 欧阳斌, 等. 基于高斯混合模型 Varian NovalisTX 直线加速器盆腔肿瘤放疗摆位误差分布预测模型的构建[ J]. 中山大学学报(医学科学版), 2019, 40 ( 2):284-290.

Qiu MM, Zhong JJ, Ouyang B, et al. Set-up errors distribution prediction model for pelvic tumors radiotherapy of Varian NovalisTX medical linear accelerator based on gaussian mixtures [ J]. Journal of Sun Yat-sen University ( Medical Sciences), 2019, 40(2):284-290.

[11] Bengio Y. Learning deep architectures for AI [ J]. Foundations and Trends ? in Machine Learning, 2009, 2(1):1-127.

[12] 张方圆, 郁芸, 赵宇, 等. 人工神经网络在临床医学中的应用[J]. 北京生物医学工程, 2016, 35(4):422-428.

Zhang FY, Yu Y, Zhao Y, et al. Application of artificial neural network in clinical medicine [ J ]. Beijing Biomedical Engineering, 2016, 35(4):422-428.

[13] Spencer M, Eickholt J, Cheng J. A deep learning network approach to ab initio protein secondary structure prediction[ J]. IEEE / ACM Transactions on Computational Biology & Bioinformatics, 2015, 12(1):103-112.

[14] 李伟, 杨金才, 黄牛. 深度学习在药物设计与发现中的应用[J]. 药学学报, 2019, 54(5):761-767.

Li W, Yang JC, Huang N. Deep learning in drug design and discovery [ J ]. Acta Pharmaceutica Sinica, 2019, 54 ( 5 ):761-767.

[15] Ayachi R, Afif M, Said Y, et al. Traffic signs detection for real-world application of an advanced driving assisting system using deep learning [ J]. Neural Processing Letters, 2020, 51 ( 1): 837-851.

[16] Sezer OB, Ozbayoglu AM. Algorithmic financial trading with deep convolutional neural networks: time series to image conversion approach [ J]. Applied Soft Computing, 2018, 70: 525-538.

[17] 邓金城, 彭应林, 刘常春, 等. 深度卷积神经网络在放射治疗计划图像分割中的应用[ J]. 中国医学物理学杂志, 2018,35(6):621-627.

Deng JC, Peng YL, Liu CC, et al. Application of deep convolution neural network in radiotherapy planning image segmentation[J].Chinese Journal of Medical Physics, 2018, 35 (6):621-627.

[18] 王燃. 基于深度学习的胸腹部肿瘤呼吸运动的实时跟踪方法研究[D]. 深圳:中国科学院深圳先进技术研究院, 2019.

Wang R. Deep learning in real-time tracking method for respiratory motion of thoracic and abdominal tumors [ D ]. Shenzhen:Shenzhen Institutes of Advanced Technology, Chinese Academy of Sciences, 2019.

[19] Dinkla AM, Wolterink JM, Maspero M, et al. MR-only brain radiation therapy:dosimetric evaluation of synthetic CTs generated by a dilated convolutional neural network [ J ]. International Journal of Radiation Oncology*Biology*Physics, 2018, 102 (4):801-812.

[20] Ibragimov B, Toesca DAS, Yuan Y, et al. Neural networks for deep radiotherapy dose analysis and prediction of liver SBRT outcomes [ J ]. IEEE Journal of Biomedical and Health Informatics, 2019, 23(5):1821-1833.

[21] Keras. The Python deep learning library [EB / OL]. (2019-11-6)[2019-11-15] https: / / keras.io / .

[22] 覃光华, 丁晶, 陈彬兵. 预防过拟合现象的人工神经网络训练策略及其应用[J]. 长江科学院院报, 2002, 19(3):59-61.

Qin GH, Ding J, Chen BB. Learning strategies of artificial neural networks for preventing over-training and their application [ J]. Journal of Yangtze River Scientific Research Institute, 2002, 19 (3):59-61.

[23] Lin S, Pan J, Han L, et al. Nasopharyngeal carcinoma treated with reduced-volume intensity-modulated radiation therapy:report on the 3-year outcome of a prospective series[ J]. International Journal of Radiation Oncology*Biology*Physics, 2009, 75 (4):1071-1078.

服务与反馈：

【文章下载】【加入收藏】

提示：您还未登录，请登录！点此登录