Prediction of Results of Radiotherapy With Ku70 Expression and an Artificial Neural Network
Prediction of Results of Radiotherapy With Ku70 Expression and an Artificial Neural Network
复制标题
DOI：
10.21873/invivo.12114
复制
发表时间：
2020-09-01
期刊：
影响因子：
2.3
通讯作者：
Sakata, Koh-Ichi
中科院分区：
文献类型：
--
作者：
Hasegawa, Tomokazu;Someya, Masanori;Sakata, Koh-Ichi
背景/目的：准确预测放射治疗效果对于肿瘤治疗方式的个体化选择是必不可少的。我们研究了人工神经网络（ANN）模型在预测放射治疗结果中的应用，使用临床因素和Ku 70免疫组化染色作为输入。患者和方法：我们分析了2001年8月至2010年10月期间接受放射治疗的79例前列腺癌患者。我们还分析了2002年3月至2009年12月间接受放射治疗的46例下咽癌伴鳞状细胞癌患者。使用标准前馈、反向传播神经网络的适当训练的ANN分析用于预测放射治疗结果。结果如下：前列腺癌调强放疗（IMRT）+雄激素剥夺治疗（ADT）患者的受试者工作特征曲线下面积（AUC）为0.939，单独IMRT为0.803，单独3D适形放疗（CRT）为0.960。IMRT+ADT的敏感性和特异性分别为85.7%和90.4%，单独IMRT的敏感性和特异性分别为75.0%和88.5%，单独3D-CRT的敏感性和特异性分别为92.3%和100%。下咽癌的AUC为0.901。敏感性和特异性分别为66.7%和88.2%。结论：我们证明了使用人工神经网络结合Ku 70表达和临床因素作为输入来预测前列腺癌和下咽癌的放射治疗结果的可能性。
Background/Aim: Accurate prediction of radiotherapy results is indispensable for the individualized selection of treatment modalities of cancer. We examined the application of the artificial neural network (ANN) model in predicting radiotherapy results using clinical factors and immunohistochemical staining of Ku70 as inputs. Patients and Methods: We analyzed 79 prostate cancer patients with localized adenocarcinoma treated with radiotherapy between August 2001 and October 2010. We also analyzed 46 hypopharyngeal cancer patients with squamous cell carcinoma treated with radiotherapy between March 2002 and December 2009. The properly trained ANN analysis using a standard feedforward, back-propagation neural network was used to predict the radiotherapy treatment results. Results: The areas under the receiver-operating characteristic curve (AUC) were 0.939 for patients treated with intensity modulated radiotherapy (IMRT)+androgen deprivation therapy (ADT), 0.803 for IMRT alone, and 0.960 for 3D-conformal radiotherapy (CRT) alone in prostate cancer. Sensitivity and specificity were 85.7% and 90.4% for IMRT+ADT, 75.0% and 88.5% for IMRT alone, and 92.3% and 100% for 3D-CRT alone. The AUC was 0.901 for hypopharyngeal cancer. Sensitivity and specificity were 66.7% and 88.2%, respectively. Conclusion: We demonstrated a possibility to predict the radiotherapy treatment results in prostate and hypopharyngeal cancer using ANN in combination with Ku70 expression and clinical factors as inputs.