人工神经网络技术与肿瘤标志物联合检测在肺癌组织分型中的价值
作者：吴拥军 吴逸明 王丽萍 屈凌波 相秉仁
单位：吴拥军 吴逸明(河南医科大学劳动卫生学与卫生毒理学教研室 郑州 450052)；王丽萍(河南医科大学第一附属医院肿瘤科 郑州 450052)；屈凌波 相秉仁(中国药科大学分析计算中心 南京 210009)
关键词：肿瘤标志物；肺癌；小细胞肺癌；非小细胞肺癌；组织分型；人工神经网络
河南医科大学学报000306 摘要 目的：通过4项肿瘤标志物联合检测，运用人工神经网络技术，提高小细胞肺癌(SCLC)与非小细胞肺癌(NSCLC)正确判别率。方法: 用放射免疫法测定了51例肺癌患者血清癌胚抗原(CEA)、糖类抗原125(CA125)、促胃液素、神经元特异性烯醇化酶(NSE)水平，采用人工神经网络技术，探讨了4项肿瘤标志物在肺癌组织分型中的应用价值。结果:SCLC患者促胃液素、NSE水平明显高于NSCLC患者，而CEA、CA125水平却低于非小细胞肺癌患者。人工神经网络技术在判别SCLC与NSCLC类型中，总的符合率为87.5%。结论:该4项肿瘤标志物联合检测在肺癌组织分型方面可为临床提供有价值的参考资料，同时表明人工神经网络技术在肺癌组织分型中具有一定的实用价值。
, 百拇医药
分类号 R734.2
Clinical value of determination of tumor markers in patients with lung carcinoma for histological type based on artificial neural network
WU Yongjun , WU Yiming ,(Department of Occupational Health and Health Toxicology Disease,Henan Medical University, Zhengzhou 450052 P.R.China)
WANG Liping ,(Department of Oncology, the First Affiliated Hospital, Henan Medical University, Zhengzhou 450052 P.R.China)
, 百拇医药
QU Lingbo, XIANG Bingren
(Center of Analysis and Computer, China Pharmaceutical University, Nanjing 210009 P.R.China
Abstract Aim: The early diagnosis of lung cancer is very important, but this is still difficult. The aim of this study was to improve the rate of correct classification of small cell and non small cell lung cancer. Methods: A panel of four tumor biomarkers, including serum CEA、CA125、gastrin and NSE were determined with radioimmunoassay in 30 patients with non small cell lung cancer (NSCLC) and 21 patients with small cell lung cancer (SCLC). We explored the usefulness of artificial neural network (ANN) in this discrimination. Results: The concentrations of CEA 、CA125、gastrin and NSE had significantly difference between the two groups. The levels of gastrin and NSE in SCLC were significantly higher than those in NSCLC, but levels of CEA andCA125 in SCLC were significantly lower than those in NSCLC. ANN was able to correctly identify all but two cases. The total accurate rate was 87.5% in distinguishing SCLC from NSCLC. Conclusions: The results revealed that CEA and CA125 had specificity in NSCLC, gastrin and NSE in SCLC. And so combined determination of the four tumor markers may be useful in the histological type diagnosis of lung cancer before operation. ANN was useful method in the prediction of histological type of lung cancer.
, 百拇医药
In recent years, lung cancer is one of the most frequent cancer in the world, and today its incidence is rapidly rising. It is forecasted that lung cancer in the 21 century is the leading disease in the epidemiology of cancer. It has been shown that cigarette is an important factor of causing lung cancer and that smoking cessation can reduce the incidences of lung cancer. Secondly, urbanization along with atmospheric pollution, perhaps is the cause of high incidence of lung cancer in the city. Other factors such as: personal character，social relationships, stimulated experience. In the case, these factors may complicate, act together and so cause lung cancer. Lung cancer prognosis is usually very bad, and a 5-year survival rate is under 50%. Lung cancer is classified into small cell and non-small cell lung cancer. NSCLC comprises 70%～80% of all lung cancer, mainly including squamous cell cancer and adenocarcinoma. SCLC is different from NSCLC in treatment and prognosis; SCLC has neuroendocrine biological properties usually associated with a high sensitivity to radio and chemotherapy[1,2]. Surgery offers the best probability of cure for NSCLC, but it cannot be proposed for locally advanced and metastasis stages, therefore, other treatments such as radiotherapy and chemotherapy combination have been proposed[3]. So the treatment scheme changes with different histological types[4]. And it is important to make certain histological type for deciding treatment scheme. For early clinical diagnosis of lung cancer, various tumor markers have been investigated, such as neuron specific enolase, gastrin, for SCLC[5, 6], and carcinoembryonic antigen, carbohydrate antigen CA125, for NSCLC[7, 8]. But for measurement of single marker, its sensitivity and specificity are hard to satisfy to clinical request for early and distinguishing diagnosis. From the point of view, combined determination proved to be more powerful than single. To improve the rate of correct classification of small-cell lung cancer and non-small-cell lung cancer, a panel of four tumor markers, including serum CEA、CA125、gastrin and NSE were determined with radioimmunoassay in 30 patients with NSCLC and 21patients with SCLC. And we explored the usefulness of artificial neural network (ANN) in this discrimination.
, 百拇医药
1 Materials and methods
1.1 Patients We used a previously characterized cohort of 51 proven patients with lung cancer treated between December 1994 and December 1997 at the First Affiliated Hospital of Henan Medical University. In these patients ,39 males and 12 females; age from 45 to 69,mean age of 60 years;15 with malignant pleural effusion,36 without pleural effusion ; 35 smokers,16 non-smokers;16 in TNM stage II, 16 in stage III,19 in stage Ⅳ.Histological classification according to World Heath Organization(WHO)criteria. Among them, NSCLC group consisted of 30 cases, including 12 squamous cell cancer and 18 adenocarcinoma. SCLC group consisted of 21 cases. In all these subjects , a blood sample was taken from each patient at presentation when he was limosis in the morning, then serum was separated and stored at -20 ℃ until test.
, http://www.100md.com
1.2 Method Serum CEA、CA125、gastrin and NSE levels were determined by radioimmunoassay according to the indications of the producers (table 1)
Table 1 Tumor Markers Marker
Producer
Assay
CEA
Jiaozuo Institute of immune reagent, ,Henan, China
RIA
CA125
, 百拇医药
tumor Institute, Chinese Academy of Medical Sciences
RIA
Gastrin
isotope Institute, Chinese Academy of Atom Energy Sciences
RIA
NSE
tumor Institute, Chinese Academy of Medical Sciences
RIA
Note: CEA: carcinoembryonic antigen; CA125: carbohydrate antigen125; NSE: neuron specific enolase; RIA: radioimmunoassay
, 百拇医药
1.3 Statistics The serum tumor marker was not distributed normally; thus, for each patient subset, results were expressed as mean value (
), and variation was expressed as standard deviation (s). Data of NSCLC and SCLC were expressed as mean value ± standard deviation. Experiment data of the two groups were tested by D test , most of the data accorded with normal distribution, if not after transformed by logarithm, all of them accorded with normal distribution. Differences between two independent groups were determined by means of t test and square difference analysis. When P<0.05 it was considered significant. The classification of SCLC and NSCLC was investigated by use of Artificial Neural Network.
, 百拇医药
2 Results
2.1 Tumor Marker Distribution The levels of tumor markers in 51 patients with lung cancer are shown in Table 2. From the data, we can see that there were notable difference between SCLC and NSCLC in the levels of CEA、CA125、gastrin and NSE. The levels of serum CEA and CA125 were significantly higher in NSCLC than those in SCLC, but the levels of serum Gastrin and NSE were significantly lower in NSCLC.
Table 2 Levels of serum CEA、CA125、Gastrin、NSE(±s) Species
, http://www.100md.com
n
ρ(CEA)/μg.L-1
ρ(CA125)/mg.L-1
ρ(Gastrin)/ng.L-1
ρ(NSE)/μg.L-1
NSCLC
30
34.09±7.08
, http://www.100md.com
71.69±52.85
138.68±39.73
36.40±17.89
SCLC
21
25.49±6.51
32.03±17.13
227.27±70.90
51.33±28.35
P
<0.001
<0.01
, 百拇医药
<0.001
<0.05
2.2 Artificial Neural Network Artificial neural network is the kind of new information disposal technique. We reported the use of artificial neural network (ANN) with back-propagation of error for developing nonlinear calibration models for tumor marker assay. The theory of ANN and arithmetic can be seen[9][10].In our study, The ANN consisted of three layers of nodes: an input layer, a hidden layer, and an output layer. In calibration applications, the input signals were used to determine values of four tumor markers in two groups, and only one linear output node was used in the output layer. The expectation output value was set at 0.3 for NSCLC, at 0.7 for SCLC. When the number of neurons in hidden layer were 1,2 and 3, the results of standard error of prediction were same, so we selected 1 as the number of neurons in hidden layer. Then we investigated effect of learning speed to standard error of prediction, and the results showed that learning speed had little effect from 0.5 to 2.0, but it could affect convergence speed. According to our study, we selected 1.5 as learning speed. During the training process, the samples of training and prediction sets were needed to be standardized to adjust sigmoid function, when desired level of precision was obtained in the estimated class, the training process stopped. The most excited aspect of this method was that the training process to accommodate nonlinear response in SCLC and NSCLC can automatically adjust the weights, it did not need to know the mean values of SCLC and NSCLC, and could distinguish the two classes.
, 百拇医药
2.3 Predictive classification for prediction samples
In a previous optimized condition, the results of predictive classification for independent prediction samples by ANN.
As shown in Fig.1, ANN was able to correctly identify all but two cases. The total accurate rate was 87.5% in distinguishing SCLC from NSCLC. The approach based on ANN proved to be very effective and simple. It could be used to provide the useful information about histological type of lung cancer before operation.
, 百拇医药
Fig1 predictive classification for prediction samples by ANN
(■symbolizecd SCLC ●symbolizecd NSCLC)
3 Discussion
Tumor markers are substances that can often be detected in higher-than-normal amounts in the blood of some patients with certain types of cancer. Tumor markers are produced either by the tumor itself or by the body in response to the presence of cancer. Carcinoembryonic antigen (CEA) is normally found in small amounts in the blood of most healthy people, but may become elevated in people who have cancer. The primary use of CEA is in monitoring colorectal cancer, but for lung cancer, CEA is considered to be good tumor marker for NSCLC.CA125 is produced by a variety of cells, but particularly by ovarian cancer cells. Some studies have shown that many women with ovarian cancer have elevated CA125 levels. But not all women with elevated CA125 levels have ovarian cancer. CA125 levels may also be elevated by lung cancer; its masculine rate is about 40% for lung cancer. According to our studies, the serum CEA、CA125 levels of NSCLC were significantly higher than those of SCLC group. Gastrin is a kind of hormone excreted by intestine and stomach. It is reported that serum gastrin concentration with lung cancer is concerned with histological type and stages[11]. Neuron-specific enolase (NSE) has been detected in patient with SCLC, studies of NSE as a tumor marker have concentrated primarily on patients with SCLC. Measurement of NSE level in patients with SCLC can provide information about extent of the disease and the patient prognosis, as well as about the response to treatment. In this paper, we could draw the same conclusion, that is, the serum gastrin、NSE levels of SCLC were significantly higher than those of NSCLC group. These results indicated that the serum CEA、CA125 levels were useful in NSCLC, and gastrin and NSE in SCLC. The assay of serum CEA、CA125 、gastrin and NSE had a certain practical value for the histological type and prognosis of lung cancer. But if single marker level was not satisfactory in distinguishing SCLC from NSCLC, because their correct rate was not high. So we utilized combined determination, using ANN developed a classificator. The diagnostic value of the combined tumor markers was much better than that of the single marker. The rate of correct classification of SCLC vs. NSCLC was 87.5%. Our detailed analysis showed that ANN improved diagnostic accuracy up to a rate of 10% and it did not need to know the mean value of each group. It proved to be more powerful than measurement of single marker and showed simple, rapid and direct. Its use may help in distinguishing SCLC from NSCLC and make it possible to define different subgroups in the earlier course of lung cancer.
, 百拇医药
基金项目：河南省自然科学基金资助项目 991170215
研究方向：肺癌的病因学、预防、早期诊断和综合治疗，作者简介：吴逸明，男，55岁，教授，博士生导师，References
[1]Gronowitz JS, Bergstrom R, Nou E, et al. Clinical and serologic markers of stage and prognosis in small cell lung cancer. Cancer, 1990,66(4): 722
[2]Carney DN. Lung cancer biology. Eur J Cancer, 1991,27(3): 366
[3]Mulshine JL, Glatstein E, Ruckdeschel JC, Treatment of non-small lung cancer. J Clin Oncol, 1986, 4:1 704
, http://www.100md.com
[4]Wilson JD. Harrison's principles of internal medicine,12th ed. New York: McGraw-Hill, 1991.1 102
[5]Akoun GM, Scarna HM,Milleron BJ, et al. Serum neuron-specific enolase: a marker for disease extent and response for therapy for small cell lung cancer. Chest, 1985,87:39
[6]Pinson P, Watripont P, Joos G, et al.Serum neuronspecific enolase as a tumor marker in the diagnosis and follow-up of small cell lung cancer. Respiration, 1997,64(1): l102
, http://www.100md.com
[7]Diez M,Pollan M,Maestro ML,et al.Concordance between serum and cytosolic level of CEA,CA125 and SCC antigens in patients with non-small cell lung cancer. Anticancer Res, 1995,15(6B): 2 811
[8]Manuel D，Antonio T，Marina P,et al. Prognostic significance of serum CA125 antigen assay in patients with non-small cell lung cancer. Cancer, 1994, 73:1 368
[9]Paul J, Gemperline, James R, et al. Nonlinear Multivariate Calibration Using Principal Components Regression and Artificial Neural Networks. Anal Chem, 1991, 63(20): 2 313
, 百拇医药
[10]Zhang Zhuoyong, Liu Sidong, Ding Baojun, et al. Artificial neural network applied to diagnosis of lung cancer. Chemical Journal of Chinese Universities,1998, 19(4):530
[11]Zhou JJ, Yun D. Clinical value of measurement of serum neuron specific enolase and gastrin for lung cancer. Cancer Research on Prevention and Treatment,1999,26(5):339
2000-01-05收稿, 百拇医药
单位：吴拥军 吴逸明(河南医科大学劳动卫生学与卫生毒理学教研室 郑州 450052)；王丽萍(河南医科大学第一附属医院肿瘤科 郑州 450052)；屈凌波 相秉仁(中国药科大学分析计算中心 南京 210009)
关键词：肿瘤标志物；肺癌；小细胞肺癌；非小细胞肺癌；组织分型；人工神经网络
河南医科大学学报000306 摘要 目的：通过4项肿瘤标志物联合检测，运用人工神经网络技术，提高小细胞肺癌(SCLC)与非小细胞肺癌(NSCLC)正确判别率。方法: 用放射免疫法测定了51例肺癌患者血清癌胚抗原(CEA)、糖类抗原125(CA125)、促胃液素、神经元特异性烯醇化酶(NSE)水平，采用人工神经网络技术，探讨了4项肿瘤标志物在肺癌组织分型中的应用价值。结果:SCLC患者促胃液素、NSE水平明显高于NSCLC患者，而CEA、CA125水平却低于非小细胞肺癌患者。人工神经网络技术在判别SCLC与NSCLC类型中，总的符合率为87.5%。结论:该4项肿瘤标志物联合检测在肺癌组织分型方面可为临床提供有价值的参考资料，同时表明人工神经网络技术在肺癌组织分型中具有一定的实用价值。
, 百拇医药
分类号 R734.2
Clinical value of determination of tumor markers in patients with lung carcinoma for histological type based on artificial neural network
WU Yongjun , WU Yiming ,(Department of Occupational Health and Health Toxicology Disease,Henan Medical University, Zhengzhou 450052 P.R.China)
WANG Liping ,(Department of Oncology, the First Affiliated Hospital, Henan Medical University, Zhengzhou 450052 P.R.China)
, 百拇医药
QU Lingbo, XIANG Bingren
(Center of Analysis and Computer, China Pharmaceutical University, Nanjing 210009 P.R.China
Abstract Aim: The early diagnosis of lung cancer is very important, but this is still difficult. The aim of this study was to improve the rate of correct classification of small cell and non small cell lung cancer. Methods: A panel of four tumor biomarkers, including serum CEA、CA125、gastrin and NSE were determined with radioimmunoassay in 30 patients with non small cell lung cancer (NSCLC) and 21 patients with small cell lung cancer (SCLC). We explored the usefulness of artificial neural network (ANN) in this discrimination. Results: The concentrations of CEA 、CA125、gastrin and NSE had significantly difference between the two groups. The levels of gastrin and NSE in SCLC were significantly higher than those in NSCLC, but levels of CEA andCA125 in SCLC were significantly lower than those in NSCLC. ANN was able to correctly identify all but two cases. The total accurate rate was 87.5% in distinguishing SCLC from NSCLC. Conclusions: The results revealed that CEA and CA125 had specificity in NSCLC, gastrin and NSE in SCLC. And so combined determination of the four tumor markers may be useful in the histological type diagnosis of lung cancer before operation. ANN was useful method in the prediction of histological type of lung cancer.
, 百拇医药
In recent years, lung cancer is one of the most frequent cancer in the world, and today its incidence is rapidly rising. It is forecasted that lung cancer in the 21 century is the leading disease in the epidemiology of cancer. It has been shown that cigarette is an important factor of causing lung cancer and that smoking cessation can reduce the incidences of lung cancer. Secondly, urbanization along with atmospheric pollution, perhaps is the cause of high incidence of lung cancer in the city. Other factors such as: personal character，social relationships, stimulated experience. In the case, these factors may complicate, act together and so cause lung cancer. Lung cancer prognosis is usually very bad, and a 5-year survival rate is under 50%. Lung cancer is classified into small cell and non-small cell lung cancer. NSCLC comprises 70%～80% of all lung cancer, mainly including squamous cell cancer and adenocarcinoma. SCLC is different from NSCLC in treatment and prognosis; SCLC has neuroendocrine biological properties usually associated with a high sensitivity to radio and chemotherapy[1,2]. Surgery offers the best probability of cure for NSCLC, but it cannot be proposed for locally advanced and metastasis stages, therefore, other treatments such as radiotherapy and chemotherapy combination have been proposed[3]. So the treatment scheme changes with different histological types[4]. And it is important to make certain histological type for deciding treatment scheme. For early clinical diagnosis of lung cancer, various tumor markers have been investigated, such as neuron specific enolase, gastrin, for SCLC[5, 6], and carcinoembryonic antigen, carbohydrate antigen CA125, for NSCLC[7, 8]. But for measurement of single marker, its sensitivity and specificity are hard to satisfy to clinical request for early and distinguishing diagnosis. From the point of view, combined determination proved to be more powerful than single. To improve the rate of correct classification of small-cell lung cancer and non-small-cell lung cancer, a panel of four tumor markers, including serum CEA、CA125、gastrin and NSE were determined with radioimmunoassay in 30 patients with NSCLC and 21patients with SCLC. And we explored the usefulness of artificial neural network (ANN) in this discrimination.
, 百拇医药
1 Materials and methods
1.1 Patients We used a previously characterized cohort of 51 proven patients with lung cancer treated between December 1994 and December 1997 at the First Affiliated Hospital of Henan Medical University. In these patients ,39 males and 12 females; age from 45 to 69,mean age of 60 years;15 with malignant pleural effusion,36 without pleural effusion ; 35 smokers,16 non-smokers;16 in TNM stage II, 16 in stage III,19 in stage Ⅳ.Histological classification according to World Heath Organization(WHO)criteria. Among them, NSCLC group consisted of 30 cases, including 12 squamous cell cancer and 18 adenocarcinoma. SCLC group consisted of 21 cases. In all these subjects , a blood sample was taken from each patient at presentation when he was limosis in the morning, then serum was separated and stored at -20 ℃ until test.
, http://www.100md.com
1.2 Method Serum CEA、CA125、gastrin and NSE levels were determined by radioimmunoassay according to the indications of the producers (table 1)
Table 1 Tumor Markers Marker
Producer
Assay
CEA
Jiaozuo Institute of immune reagent, ,Henan, China
RIA
CA125
, 百拇医药
tumor Institute, Chinese Academy of Medical Sciences
RIA
Gastrin
isotope Institute, Chinese Academy of Atom Energy Sciences
RIA
NSE
tumor Institute, Chinese Academy of Medical Sciences
RIA
Note: CEA: carcinoembryonic antigen; CA125: carbohydrate antigen125; NSE: neuron specific enolase; RIA: radioimmunoassay
, 百拇医药
1.3 Statistics The serum tumor marker was not distributed normally; thus, for each patient subset, results were expressed as mean value (
), and variation was expressed as standard deviation (s). Data of NSCLC and SCLC were expressed as mean value ± standard deviation. Experiment data of the two groups were tested by D test , most of the data accorded with normal distribution, if not after transformed by logarithm, all of them accorded with normal distribution. Differences between two independent groups were determined by means of t test and square difference analysis. When P<0.05 it was considered significant. The classification of SCLC and NSCLC was investigated by use of Artificial Neural Network., 百拇医药
2 Results
2.1 Tumor Marker Distribution The levels of tumor markers in 51 patients with lung cancer are shown in Table 2. From the data, we can see that there were notable difference between SCLC and NSCLC in the levels of CEA、CA125、gastrin and NSE. The levels of serum CEA and CA125 were significantly higher in NSCLC than those in SCLC, but the levels of serum Gastrin and NSE were significantly lower in NSCLC.
Table 2 Levels of serum CEA、CA125、Gastrin、NSE(
±s) Species, http://www.100md.com
n
ρ(CEA)/μg.L-1
ρ(CA125)/mg.L-1
ρ(Gastrin)/ng.L-1
ρ(NSE)/μg.L-1
NSCLC
30
34.09±7.08
, http://www.100md.com
71.69±52.85
138.68±39.73
36.40±17.89
SCLC
21
25.49±6.51
32.03±17.13
227.27±70.90
51.33±28.35
P
<0.001
<0.01
, 百拇医药
<0.001
<0.05
2.2 Artificial Neural Network Artificial neural network is the kind of new information disposal technique. We reported the use of artificial neural network (ANN) with back-propagation of error for developing nonlinear calibration models for tumor marker assay. The theory of ANN and arithmetic can be seen[9][10].In our study, The ANN consisted of three layers of nodes: an input layer, a hidden layer, and an output layer. In calibration applications, the input signals were used to determine values of four tumor markers in two groups, and only one linear output node was used in the output layer. The expectation output value was set at 0.3 for NSCLC, at 0.7 for SCLC. When the number of neurons in hidden layer were 1,2 and 3, the results of standard error of prediction were same, so we selected 1 as the number of neurons in hidden layer. Then we investigated effect of learning speed to standard error of prediction, and the results showed that learning speed had little effect from 0.5 to 2.0, but it could affect convergence speed. According to our study, we selected 1.5 as learning speed. During the training process, the samples of training and prediction sets were needed to be standardized to adjust sigmoid function, when desired level of precision was obtained in the estimated class, the training process stopped. The most excited aspect of this method was that the training process to accommodate nonlinear response in SCLC and NSCLC can automatically adjust the weights, it did not need to know the mean values of SCLC and NSCLC, and could distinguish the two classes.
, 百拇医药
2.3 Predictive classification for prediction samples
In a previous optimized condition, the results of predictive classification for independent prediction samples by ANN.
As shown in Fig.1, ANN was able to correctly identify all but two cases. The total accurate rate was 87.5% in distinguishing SCLC from NSCLC. The approach based on ANN proved to be very effective and simple. It could be used to provide the useful information about histological type of lung cancer before operation.
, 百拇医药
Fig1 predictive classification for prediction samples by ANN
(■symbolizecd SCLC ●symbolizecd NSCLC)
3 Discussion
Tumor markers are substances that can often be detected in higher-than-normal amounts in the blood of some patients with certain types of cancer. Tumor markers are produced either by the tumor itself or by the body in response to the presence of cancer. Carcinoembryonic antigen (CEA) is normally found in small amounts in the blood of most healthy people, but may become elevated in people who have cancer. The primary use of CEA is in monitoring colorectal cancer, but for lung cancer, CEA is considered to be good tumor marker for NSCLC.CA125 is produced by a variety of cells, but particularly by ovarian cancer cells. Some studies have shown that many women with ovarian cancer have elevated CA125 levels. But not all women with elevated CA125 levels have ovarian cancer. CA125 levels may also be elevated by lung cancer; its masculine rate is about 40% for lung cancer. According to our studies, the serum CEA、CA125 levels of NSCLC were significantly higher than those of SCLC group. Gastrin is a kind of hormone excreted by intestine and stomach. It is reported that serum gastrin concentration with lung cancer is concerned with histological type and stages[11]. Neuron-specific enolase (NSE) has been detected in patient with SCLC, studies of NSE as a tumor marker have concentrated primarily on patients with SCLC. Measurement of NSE level in patients with SCLC can provide information about extent of the disease and the patient prognosis, as well as about the response to treatment. In this paper, we could draw the same conclusion, that is, the serum gastrin、NSE levels of SCLC were significantly higher than those of NSCLC group. These results indicated that the serum CEA、CA125 levels were useful in NSCLC, and gastrin and NSE in SCLC. The assay of serum CEA、CA125 、gastrin and NSE had a certain practical value for the histological type and prognosis of lung cancer. But if single marker level was not satisfactory in distinguishing SCLC from NSCLC, because their correct rate was not high. So we utilized combined determination, using ANN developed a classificator. The diagnostic value of the combined tumor markers was much better than that of the single marker. The rate of correct classification of SCLC vs. NSCLC was 87.5%. Our detailed analysis showed that ANN improved diagnostic accuracy up to a rate of 10% and it did not need to know the mean value of each group. It proved to be more powerful than measurement of single marker and showed simple, rapid and direct. Its use may help in distinguishing SCLC from NSCLC and make it possible to define different subgroups in the earlier course of lung cancer.
, 百拇医药
基金项目：河南省自然科学基金资助项目 991170215
研究方向：肺癌的病因学、预防、早期诊断和综合治疗，作者简介：吴逸明，男，55岁，教授，博士生导师，References
[1]Gronowitz JS, Bergstrom R, Nou E, et al. Clinical and serologic markers of stage and prognosis in small cell lung cancer. Cancer, 1990,66(4): 722
[2]Carney DN. Lung cancer biology. Eur J Cancer, 1991,27(3): 366
[3]Mulshine JL, Glatstein E, Ruckdeschel JC, Treatment of non-small lung cancer. J Clin Oncol, 1986, 4:1 704
, http://www.100md.com
[4]Wilson JD. Harrison's principles of internal medicine,12th ed. New York: McGraw-Hill, 1991.1 102
[5]Akoun GM, Scarna HM,Milleron BJ, et al. Serum neuron-specific enolase: a marker for disease extent and response for therapy for small cell lung cancer. Chest, 1985,87:39
[6]Pinson P, Watripont P, Joos G, et al.Serum neuronspecific enolase as a tumor marker in the diagnosis and follow-up of small cell lung cancer. Respiration, 1997,64(1): l102
, http://www.100md.com
[7]Diez M,Pollan M,Maestro ML,et al.Concordance between serum and cytosolic level of CEA,CA125 and SCC antigens in patients with non-small cell lung cancer. Anticancer Res, 1995,15(6B): 2 811
[8]Manuel D，Antonio T，Marina P,et al. Prognostic significance of serum CA125 antigen assay in patients with non-small cell lung cancer. Cancer, 1994, 73:1 368
[9]Paul J, Gemperline, James R, et al. Nonlinear Multivariate Calibration Using Principal Components Regression and Artificial Neural Networks. Anal Chem, 1991, 63(20): 2 313
, 百拇医药
[10]Zhang Zhuoyong, Liu Sidong, Ding Baojun, et al. Artificial neural network applied to diagnosis of lung cancer. Chemical Journal of Chinese Universities,1998, 19(4):530
[11]Zhou JJ, Yun D. Clinical value of measurement of serum neuron specific enolase and gastrin for lung cancer. Cancer Research on Prevention and Treatment,1999,26(5):339
2000-01-05收稿, 百拇医药