| 尹柯,巴文娟,汤敏,刘金沙,汪琼,孙婷婷,谢梅,沈晶,林琳,伍建林.多层感知器模型预测纯磨玻璃结节肺腺癌浸润程度[J].中国医学影像技术,2020,36(11):1652~1656 |
| 多层感知器模型预测纯磨玻璃结节肺腺癌浸润程度 |
| Multilayer perceptron model in predicting infiltration degree of pure ground glass opacity lung adenocarcinoma |
| 投稿时间：2019-12-23 修订日期：2020-08-23 |
| DOI：10.13929/j.issn.1003-3289.2020.11.014 |
| 中文关键词: 肺腺癌 神经网络,计算机 体层摄影术,X线计算机 |
| 英文关键词:adenocarcinoma of lung neural networks, computer tomography, X-ray computed |
| 基金项目: |
| 作者 | 单位 | E-mail | | 尹柯 | 大连大学附属中山医院放射科, 辽宁 大连 116000 | | | 巴文娟 | 大连大学附属中山医院放射科, 辽宁 大连 116000 | | | 汤敏 | 大连大学附属中山医院放射科, 辽宁 大连 116000 | | | 刘金沙 | 大连大学附属中山医院放射科, 辽宁 大连 116000 | | | 汪琼 | 大连大学附属中山医院放射科, 辽宁 大连 116000 | | | 孙婷婷 | 大连大学附属中山医院放射科, 辽宁 大连 116000 | | | 谢梅 | 大连大学附属中山医院放射科, 辽宁 大连 116000 | | | 沈晶 | 大连大学附属中山医院放射科, 辽宁 大连 116000 | | | 林琳 | 大连大学附属中山医院放射科, 辽宁 大连 116000 | | | 伍建林 | 大连大学附属中山医院放射科, 辽宁 大连 116000 | cjr.wujianlin@vip.163.com |
|
| 摘要点击次数: 2398 |
| 全文下载次数: 713 |
| 中文摘要: |
| 目的 利用CT数据建立预测纯磨玻璃结节（pGGN）肺腺癌浸润程度的多层感知器（MLP）模型，并验证其诊断效能。方法 收集2015年1月—2018年10月393例接受手术治疗并经术后病理证实为肺腺癌或不典型腺瘤样增生（AAH）的pGGN患者（共442枚pGGN）作为训练集，建立MLP模型和二元Logistic回归模型。以2019年6月―8月89例接受手术治疗的pGGN患者（共100枚pGGN）作为验证集，利用受试者工作特征（ROC）曲线下面积（AUC）及模型预测准确率、敏感度及特异度评估其效能。结果 二元Logistic回归模型验证集的AUC、预测准确率、敏感度及特异度分别为0.799、0.820、0.683及0.915，而MLP模型验证集分别为0.869、0.880、0.805及0.932，MLP模型较二元Logistic回归模型绝对净重新分类改善指数（NRI）为6%（Z=3.473、P=0.001）。结论 所建MLP模型对预测表现为pGGN的肺腺癌中的IA具有较高准确率。 |
| 英文摘要: |
| Objective To establish a multilayer perceptron (MLP) model for predicting infiltration degree of pure ground glass nodule (pGGN) lung adenocarcinoma using CT data, and to verify its diagnostic efficiency. Methods A total of 393 pGGN patients (442 pGGN) proved to be lung adenocarcinoma or atypical adenomatoid hyperplasia (AAH) by pathology after surgical resection from January 2015 to October 2018 were collected as training set. MLP model and binary Logistic regression model were established. Other 89 pGGN patients(100pGGN) underwent surgical treatment from June to August 2019 were collected as validation set. The area under the receiver operating characteristic (ROC) curve (AUC), and the accuracy, sensitivity and specificity of the model were used to evaluate the efficacy of models. Results The AUC, prediction accuracy, sensitivity and specificity of binary Logistic regression model validation set was 0.799, 0.820, 0.683 and 0.915, respectively, while those of MLP model was 0.869, 0.880, 0.805 and 0.932, respectively. The absolute net reclassification improvement index (NRI) of MLP model was 6% (Z=3.473, P=0.001). Conclusion The established MLP model had high accuracy in predicting IA in lung adenocarcinoma presenting as pGGN. |
| 查看全文 查看/发表评论 下载PDF阅读器 |
|
|
|
