基于人工神经网络的肺癌细胞形态学诊断模型的建立及应用  

Establishment and use of a diagnostic model for morphology of lung cancer cells based on artificial neural network （ANN）

作　　者：龚平[1] 郭华雄[1] 王文清[1] 赵廷宽[1] 李春燕[1] 

机构地区：[1]华中科技大学同济医学院附属荆州医院病理科,湖北省荆州市434020

出　　处：《中华生物医学工程杂志》2012年第1期73-76,共4页Chinese Journal of Biomedical Engineering

基　　金：基金项目：荆州市医疗卫生科技发展计划（2009-13-19）

摘　　要：目的使用纤维支气管镜刷片细胞形态学定量参数建立基于人工神经网络（ANN）的诊断模型，并验证其在辅助诊断肺癌中的价值。方法利用HMIAS-2000医学图像分析系统，对组织病理学确诊的138例患者纤维支气管镜刷片细胞的细胞核进行形态定量研究，包括肺腺癌48例、肺鳞癌28例、肺小细胞癌22例，肺良性病变40例。取系统误差阈值为10^-8，随机数字法选取22例肺癌、8例肺良性病变对获得的22项参数进行ANN建模及模型训练，并用盲法测试验证模型对肺癌诊断的敏感性和特异性。结果所建立的ANN模型经过18次训练后即可达到误差要求。ANN模型诊断肺癌的敏感性为94．7％（72／76），特异性为96．9％（31／32）。结论使用纤维支气管镜刷片细胞形态学定量参数成功建立了基于ANN的诊断模型，对肺癌的鉴别诊断具有一定的应用价值。Objective To establish an ANN-based diagnostic model which uses morphology quantitative parameters of bronchoscopic brush-off cells, and to evaluate its value in adjunctive diagnosis of lung cancer. Methods By using HMIAS-2000 medical image analytical system, a quantitative morphological study was conducted in the nuclei of bronchoseopie brush-off ceils from 138 patients with histopathologically confirmed pulmonary lesions, including 48 cases of adenocareinoma, 28 of squamous carcinoma, 22 of small cell lung cancer and 40 of benign lesions. ANN modeling and training were completed by using 22 parameters obtained from random-digit selected cases of lung cancer （n=22） and pulmonary benign lesion （n=8） , with the threshold of systemic error being 10-8. A blind-test set was used to test the sensitivity and specificity of the model in diagnosing lung cancer. Results The error level of ANN model was achieved after 18 cycles of training. The ANN model had a sensitivity of 94.7%（72/76） and a specificity of 96.9% （31/32） in diagnosing lung cancer. Conclusion An ANN-based diagnostic model which uses morphology quantitative parameters of bronehoseopic brush-off cells is successfully established, and appears valuable in differential diagnosis of lung cancer.

关 键 词：支气管镜检查 细胞学 肺肿瘤 人工神经网络 诊断模型 

分 类 号：R734.2[医药卫生—肿瘤]