论文部分内容阅读
In recent years, near-infrared spectroscopy (NIRS) as a novel approach for cancer diagnosis has attracted much attention.In this study, NIR spectra of 77 endometrium sections (malignant, hyperplasia and normal samples) were firstly collected by using Nicolet 6700 extended Fourier transform near-infrared spectrometer.The original spectra were first smoothed using the Savitzky-Golay algorithm by means of a five-point cubic polynomial to reduce high-frequency noise, and multiplicative scatter correction (MSC) to correct the baseline effects.Based on the combination of fuzzy rule-building expert system (FuRES) theory and principal component analysis (PCA), a proper spectral region was selected.A FuRES model was developed to classify the different specimens based on the NIR preprocessed data of endometrial tissues.Seventy seven samples were randomly divided into calibration set (to construct FuRES models) and test set (to estimate the prediction capacity).Bootstrapped Latin-partitions method was applied to internal validations.The FuRES model with optimal components was used to classify the test set, and the satisfactory accuracy rate of 100% was obtained.The results showed that the NIRS of tissues combined with FuRES technique facilitated the classification of endometrial specimens and will improve the diagnostic approaches of endometrial carcinoma in future.(This work is supported by Natural Science Foundation of China (Grand numbers: 20875065* 30772322).