《由logistic多因素回归分析组成的“最优标志物群”结合人工神经网络用于肺癌诊断》由会员分享，可在线阅读，更多相关《由logistic多因素回归分析组成的“最优标志物群”结合人工神经网络用于肺癌诊断（2页珍藏版）》请在金锄头文库上搜索。
1、ARTIFICIAL NEURAL NETWORKS (ANN) FOR DIAGNOSIS OF LUNG CANCER : RESULTS OF “OPTIMUM TUMOR MARKERS GROUP” BY LOGISTIC REGRESSION Yongjun Wu 1, Yiming Wu 1, Lingbo Qu 1, Jing Wang 2 1 College of Public Health, Zhengzhou University, Zhengzhou, China; 450052 2 Department of Respiratory Medicine, The Fir
2、st Affiliated Hospital, Zhengzhou University, Zhengzhou, China; 450052 ABSTRACT Objectives Lung cancer is popular tumor in the world, it is estimated there are about 1.2 million people ever year. The death for lung cancer accounts for 17.8% among the whole cancer. So in the world, lung cancer is mai
3、n popular tumor in the 21 century. The early diagnosis is very important to improve survival rates of patients. Tumor markers are substances that can often be detected in higher-than-normal amounts in the blood, urine, or body tissues of some patients with certain types of cancer. It is very useful
4、and convenient for the early diagnosis. However, single marker is not sufficient to diagnose cancer, its sensitivity and specificity are not so good. To improve the diagnostic efficiency of tumor marker, Combined tumor groups are proved to be more powerful, especially used by mathematical evaluation
5、 of tumor marker profile employing artificial neural network (ANN) modeling. ANN modeling is very suitable for discrimination of lung cancer. Methods In this paper, CEA、CA125、NSE、2-MG、gastrin、 sIL-6R、 sialic acid、 pseudouridine、 NO、 Cu、 Zn、 Ca ,12 tumor markers were selected. including CEA、 CA125、 G
6、astrin and NSE , were detected by RIA 、 IRMA 、 HPLC and spectrophotometry in 50 patients with lung cancer of different histology and stage, 40 patients with benign lung diseases, 50 cases normal control group, repectively. The relevance of various indexes to lung cancer was established by logistic m
7、ultiple factors regression analysis, the general methods and suitable indexes were selected, which are the most important for diagnosis, were found to consist of “optimum tumor markers group”. According to “optimum tumor markers group”, the intelligent diagnosis system was developed based on ANN. Re
8、sults This method is superior to traditional medical statistical method. The specificity and accuracy have been remarkably improved to 28% and 21%, respectively. The rates of discrimination and predication are both 100%, The system can be used as discriminator among normal, benign and malignant. Con
9、clusions Artificial neural network has been first introduced into the diagnosis of lung cancer, and developed the intelligence diagnosis system. The system can help to improve the early diagnosis level of lung cancer, and takes on high efficiency, rapid and accuracy. The system is infavor of extensi
10、ve survey for high-risk person, and can provide effect method for clinical diagnosis. Keywords: Diagnosis of lung cancer; Tumor marker; Artificial neural network ACKNOWLEDGEMENT: This paper is assisted by sustentation fund of “211 project” in Zhengzhou University. 由 Logistic 多因素回归分析组成的“最优标志物群”结合 人工神
11、经网络用于肺癌诊断 吴拥军 1 吴逸明1 屈凌波1 王静2 1 郑州大学公共卫生学院 郑州 450052 2 郑州大学第一附属医院呼吸内科 郑州 450052 摘 要 肺癌是全球最为常见的癌症，估计每年此病的新发病例有 120 万人，肺癌占 所有癌症死亡人数中的 17.8%。因此 在全球范围内，21 世纪肺癌将是癌症流行 中的主要疾病。而早期诊断是提高肺癌生存率和降低死亡率的关键。 肿瘤标志物在肿瘤诊断、检测肿瘤的复发与转移、判断疗效和预后、群体随 防与肿瘤普查等方面都有较大的实用价值。 但是单一标志物的检测其灵敏度及特 异性均难以满足临床对早期诊断、鉴别诊断的要求，为了提高阳性率，还要寻求
12、多种标志物联合检测。 人工神经网络是理论化的人脑神经网络的数学模型，是基于模仿大脑神经网 络结构和功能而建立的一种信息处理系统。 本文采用放射免疫学、酶联免疫吸附试验、高效液相色谱法、原子吸收分光 光度法等多学科联用的手段分别测定 50 例正常对照、40 例肺良性病患者及 50 例肺癌患者血清中CEA、CA125、胃泌素、NSE、2-MG、胃泌素、sIL-6R、唾液酸、 伪尿核苷、一氧化氮、Cu、Zn及Ca的水平，探讨这些肿瘤标志物在肺癌诊断或预 警中的价值，同时应用logistic多因素回归分析法，在可能影响的多因素中“挑 选”变量，以找出对这批观测数据“最优”的变量，并分别确立各种指标对于
13、肺 癌发生的相关性，从 12 项肿瘤标志物中筛选出合适的指标，找出对肺癌诊断意 义最大的几项，组成“最优标志物群” 。在此基础上，结合人工神经网络技术， 构建出基于人工神经网络的智能化的肺癌诊断系统。从结果可以看出，人工神经 网络处理优于常规统计学方法，虽然敏感性相差不大，但特异性和准确度却有明 显差别，其结果分别比常规统计学方法提高 28%和 21%，且对肺癌诊断的识别率 和预示率均为 100%，说明用神经网络模型预测肺癌的敏感性、特异性都很好。 它不仅能够区别肺癌患者与正常人或良性患者， 而且还可以判别正常人与良性患 者。因此本研究建立的基于神经网络的智能化诊断系统，为肺癌智能诊断系统的 研制开辟了一条新途径。本系统的建立有利于高危人群大规模普查，通过本系统 不仅可以普查出肿瘤患者，而且对于良性患者也给予提示，对于早期发现、早期 预防起到积极作用。 关键词关键词 肺癌诊断；肿瘤标志物；人工神经网络； 致致 谢谢：本项目由教育部“211 工程”重点学科研究项目资助和国家自然科学基 金资助
