The diagnostic value of four serum tumor markers for gastric cancer based on pattern recognition techniques*
10.3969/j.issn.1671-8348.2017.15.015
- VernacularTitle:基于血清4项肿瘤标志物的模式识别技术对胃癌的诊断价值
- Author:
Lin GUI
;
Yuanshuai HUANG
- Keywords:
stomach neoplasms;
diagnosis;
principle component analysis;
decision tree analysis;
Fisher discriminant analysis
- From:
Chongqing Medicine
2017;46(15):2060-2062
- CountryChina
- Language:Chinese
-
Abstract:
Objective To evaluate the diagnostic value of serum tumors CA72-4,CA242,CA19-9 and carcino-embryonic antigen(CEA)in patients with gastric cancer based on pattern recognition techniques.Methods Data of serum concentrations of CA72-4,CA242,CA19-9 and CEA of 212 patients with gastric cancer,116 patients with benign gastric disease and 117 healthy subjects were retrospectively analyzed;and the diagnostic performance of each tumor marker,four tumor markers based principle component analysis(PCA),decision tree,PCA-decision tree and the fisher discriminant analysis models were established.Results CA242 had the best diagnostic effect on gastric cancer,and the area under the ROC curve(AUC)was 0.841(95％CI:0.804-0.877).PCA model showed that the serum levels of four tumor markers in patients with gastric cancer were significantly different from those in benign and healthy patients,and obvious metabolic disorders of serum with four tumor markers were found among the patients with gastric cancer.The diagnosis accuracy of the decision tree,PCA-decision tree and the Fisher discriminant analysis models for gastric cancer patients was 58.6％,65.5％and 58.6％respectively,and for non-gastric cancer patients(benign gastric diseases and healthy controls)was 94.7％,99.4％and 97.6％.And the prediction accuracy of the decision tree,PCA-decision tree and the fisher discriminant analysis models for gastric cancer patients was 65.7％,77.6％and 73.1％,and for non-gastric cancer patients was 87.5％,96.9％and 96.9％,respectively.Conclusion The PCA-decision tree model of serum CA72-4,CA242,CA19-9 and CEA might be helpful for the diagnosis and prediction of patients with gastric cancer.
