Study On Multivariate Statistical Methods Used For Fluorescence Spectrum For Stomach Cancer Diagnosis

More and more attention has been paid to fluorescence spectroscopy for its advantages of real time detection, impersonality, innocuity and so on. Although the cancer identification using spectroscopy has been studied extensively, it is still at primary stage. Especially the slow spectrum analysis technology has kept back the development of fluorescence spectrum for stomach cancer diagnosis.Light induced fluorescence was used in the thesis. 308nm, 337nm, 365nm and 405 run were chosen as excitation wavelength. The 58 normal samples and 58 cancer samples were taken from 58 stomach cancer patients. The 116 samples in vitro have been measured. The study on the fluorescence spectra revealed that 308nm and 337nm might be the optimal excitation wavelengths and they were selected for the discriminating analysis. The double factors discrimination, the partial Least-squares(PLS) method and the artificial neural network based on the partial Least-squares(ANN-PLS) were used to analyze fluorescence spectra of stomach tissues. The sensitivity, specificity and the positive predictive value of the three classification algorithms were 68.4%, 94.7%, 92.9%; 81.0%, 96.6% , 95.9%; 86.2% , 100% , 100% in separately. So we concluded that the ANN-PLS method was the most effective method for identification of stomach cancer.