| Background and ObjectionAccording to the statistical data from the Ministry of Health, the cancer mortality rate has become the leading cause of death in recent years and showed trends to be increasing in china year by year. The lung cancer, liver cancer, stomach cancer, esophageal cancer and colorectal cancer are the top five of high mortality rate. Therefore, lung cancer and gastrointestinal cancer have become a major killer of endangering China's national health. Through a great effort has been made in the prevention and treatment of tumor, the cancer morbidity and mortality still remains very high. The reasons might be not enough commitment in the tumor primary prevention and the secondary prevention, and lacking of effective community-based prevention interventions, On the other hand, the technology of tumor early diagnosis is still not available. Many cancer patients who were diagnosed at late stage have the poor treatment effect and prognosis. Detecting, diagnosing and treating at early stage are still the main measures to raise the survival rate, to reduce mortality and to improve the prognosis of cancer patients.At present, the major methods of diagnosis include imaging diagnosis, serological diagnosis, cytological and histopathologic diagnosis. However, due to the pathologist's subjective factors and the willingness of the patient to biopsy, there are many misdiagnosed cases of the cancer patients. The discovery of tumor markers makes it possible to detect tumor at early stage, however, single tumor marker is still unable to diagnose tumors at early stage. Thus, people are making hope in jointing diagnosis of tumor markers, and finding new tumor markers. These kind of effort do really make a significant improving for tumor early diagnosis. In recent years, people have done a great deal of researches to establish an intelligent diagnosis model combining the multi-tumor markers and overcome the influence of many subjective factors, which tor improve the sensitivity of tumor diagnosis.Artificial neural network (ANN) is a simulation of the human brain's organizational structure and operational mechanism. Its essence is a nonlinear information processing-system. It has a great capability of distributing and storing information, self-adjusting, self-organizing, as well as self-learning function. Intelligent diagnosis system has been widely used in disease diagnosing, screening and evaluating of risk factors, as well as gene identification and protein structure analysis. In recent years, a great effort has been paid on the tumor intelligent diagnosis system with ANN technique. Materials and MethodsThe serum samples and basic information were collected form 61 cases of normal, 53 cases of benign lung disease patients, 67 cases of lung cancer patients, 55 cases of benign stomach disease patients and 47 cases of gastric cancer patients. The serum carcinoembryonic antigen (CEA), neuron-specific enolase (NSE) and the gastrin were measured by radioimmunoassay in 283 samples; The serum calcium ion concentration was measured by Act Arsenazo III calcium determination kit in 283 samples. Serum copper and zinc ion were measured by atomic absorption spectrophotometry (graphite furnace and flame method) in 283 samples, and then calculated the ratio of copper and zinc (Cu/Zn). The serum sialic acid was measured by the modified method of resorcinol in 283 samples. Descriptive statistics, chi-square test and analysis of variance were used to analyze the data of 283 cases' basic information and six tumor markers by SPSS12.0. Each group data was normalized and divided into the training set and test set randomly. SPSS12.0 and Matlab7.0 were used to build ANN models and Logistic regression models of the lung cancer intelligent diagnosis by training set and predict the test set. Receier operating characteristic cures (ROC) were used to compare the difference between the ANN model and logistic regression model in diagnosing lung cancers and stomach cancers.Results1.Excepted for cooking style, there were different distribution with history of present illness, living habits and living environment in normal group, benign pulmonary diseases group and lung cancer group (P<0.05). These factors were closely related to the occurrence of tumor. All these could serve as risk factors for lung cancer. Collection and analysis of that essential information could also help us to determine the high risk group of lung cancer, warn and diagnose lung cancer early.2. Variance analysis and Multiple comparisons showed that the expression levels of the six tumor markers were statistically significant difference in Normal group, benign pulmonary diseases group and lung cancer group (P<0.05); Variance analysis showed that the expression levels of the six tumor markers were statistically significant difference in normal group, benign gastric diseases group and gastric cancer group (P<0.05). But multiple comparisons showed the expression levels of gastrin and CEA were no statistically significant difference between the normal group and benign diseases group of the stomach (P> 0.05).3. ANN model to diagnose lung cancer was established, according to the data of the expression with six tumor markers in normal group, benign pulmonary disease group and lung cancer group. Before and after including the basic information of three groups, The sensitivity, specificity and accuracy of predicting lung cancer in total samples using ANN model were 92.5%, 96.5%, 95.5% and 95.5%, 99.1%, 97.8%, while the sensitivity, specificity and accuracy in test set were 75.0%, 85.3%, 81.5% and 85.0%, 97.1%, 92.6%; merging with the previous experimental data, another ANN model was built by merged data. The sensitivity, specificity and accuracy of predicting lung cancer in total samples using ANN model were 94.0%, 90.4% and 91.7%, while the sensitivity, specificity and accuracy in test set were 92.0%, 98.1% and 87.3%. ANN model to diagnose gastric cancer was established, according to the data of the expression with six tumor markers in normal group, gastric benign disease group and gastric cancer group. The sensitivity, specificity and accuracy of predicting lung cancer in test set using ANN model were 88.9%, 95.7% and 93.75%; ANN model to diagnose gastric cancer and lung cancer was established, according to the data of the expression with six tumor markers in gastric cancer group and lung cancer group. The sensitivity, specificity and accuracy of predicting lung cancer in test set using ANN model were 100% and 83.3%. The sensitivity, specificity and accuracy of predicting gastric cancers in test set using ANN model were 83.3% and 100%. The overall accuracy was 93.5%.4. Before and after including basic information, area Under the ROC to evaluate the effect of lung cancer diagnosis using ANN model in total samples were 0.958 and 0.973, respectively. These were both higher than those of logistic regression models (0.958 and 0.973), and the difference between the two models was not statistically significant (P>0.05). Area Under the ROC to evaluate the effect of lung cancer diagnosis using ANN model in the test set were 0.88 and 1.0, respectively. These were both higher than those of logistic regression models (0.82 and 0.9), and the difference was also not statistically significant (P>0.05).After merging with previous research data, the ANN model's AUC of predicting total samples was 1.0. However, the logistic regression model' AUC was 1.0 too. The difference between two was not statistically significant (P>0.05). The ANN model's AUC of predicting test set was 0.95, which was greater than that of logistic regression models (0.85), and the difference between these was great statistically significant (P<0.05). The ANN model's AUC of predicting stomach cancer was 0.944, less than that of logistic regression models (1.0). The difference was no statistical significance between the two models (P>0.05). ANN model' AUC of the differential diagnosis for lung cancer and gastric cancer was 0.917, greater than those of logistic regression models (0.890). The difference was not statistically significant between the two models (P>0.05).Conclusion1. The ANN prediction model for differential diagnosis to lung cancer which based on six tumor markers had a higher sensitivity and specificity. This study confirmed that the ANN prediction model previously established had a good reproducibility and stability. Lung cancer patients' basic clinical information such as the history of present illness, family history of cancer, as well as living environment could improve the ANN prediction model's sensitivity and specificity of predicting lung cancer.2. Combined with six tumor markers, the ANN prediction model not only can identify lung cancer, benign lung disease and normal, but also can distinguish the patients with gastric cancers and benign stomach diseases, and also had a good ability to identify the patients with lung cancers and gastric cancers.3. Combined with six tumor markers, the ANN prediction model had the same predictive power of classification with logistic regression model in the small sample size. But, the ANN prediction model had an advantage in dealing with large sample size and non-linear data. |
PDF Full Text Request