Diagnosis of endometrial cancer based on near infrared spectroscopy and general regression neural network

Endometrial cancer is one of the most common cancers in women worldwide. Early stage and accurate diagnosis is indispensable for treatment of endometrial patient. In this study, near-infrared spectra of 18 normal, 30 hyperplasia, and 29 malignant pathological sections were collected. The original spectra were pretreated by using the 1st derivative Savitzky-Golay and direct orthogonal signal correction (DOSC) to improve the signal-to-noise ratio and remove the influences of background and baseline. Principal component analysis (PCA) was executed to reduce the amount of computation, and 6 principal variables were extracted as the inputs of general regression neural networks (GRNN). The spread parameters was optimized based on the RMSE of leave-one-out cross validation (LOOCV). The optimal model of GRNN built can successfully classify all the samples into three groups. The results showed that GRNN coupling with NIR spectroscopy can provide an efficient method for the early diagnosis of endometrial cancer.