[Early stage diagnosis of endometrial cancer based on near infrared spectroscopy and support vector machine]

Near-infrared spectroscopy combined with chemometrics methods for diagnosis of cancer has been reported in literatures. In our study, the NIR spectra of 77 specimens of different physiological stages of endometrium were collected. Spectral data were pretreated firstly by multiplicative scatter correction (MSC), orthogonal signal correction (OSC), and both of them, respectively, and then by SG smoothing. Latin partition method was used to select 3/4 samples as a training set, and the other 1/4 samples for test set. Support vector machine (SVM) model was built for classification, and the classification results was compared with that of partial least squares (PLS) model based on the same pretreatment methods. Samples of malignant, hyperplasia and normal endometrium were classified better by SVM (classification accuracy was 92%) than PLS (classification accuracy was 90%). The results suggested that classification accuracy was affected by pretreatment methods and models. SVM combined with endometrial tissue near infrared spectroscopy is expected to develop into a new approach to tumor diagnosis.