子宫内膜癌组织近红外光谱预处理与波段选择研究

应用近红外光谱技术对子宫内膜组织病理切片进行快速无损检测。收集了154样品光谱,其中正常样本的个数36个,增生的60个,癌变的58个。由于原始光谱中包含大量干扰信息,所以光谱预处理方法和波段选取的方法在近红外光谱分析中占有非常重要的地。利用多种预处理方法,包括一阶导数、多元散射校正、多项式最小二乘拟合求导、标准归一化、平滑、移动窗口中值滤波,对样品光谱进行了预处理。利用标准偏差谱来选取最优波段,选取的最优波段范围为4 000~6 000 cm-1。然后用处理后的光谱数据进行主成分分析,分类准确率达到100%。研究结果表明近红外光谱技术结合化学计量学方法可以作为一种癌症快速诊断的新技术,对于癌症的早期诊断和癌症组织的恶化过程研究具有重要的意义。

Study of Near Infrared Spectral Preprocessing and Wavelength Selection Methods for Endometriai Cancer Tissue

Near infrared spectroscopy was applied to measure the tissue slice of endometrial tissues for collecting the spectra. A total of 154 spectra were obtained from 154 samples. The number of normal, hyperplasia, and malignant samples was 36, 60, and 58, respectively. Original near infrared spectra are composed of many variables, for example, interference information inclu-ding instrument errors and physical effects such as particle size and light scatter. In order to reduce these influences, original spectra data should be performed with different spectral preprocessing methods to compress variables and extract useful informa-tion. So the methods of spectral preprocessing and wavelength selection have played an important role in near infrared spectros-copy technique. In the present paper the raw spectra were processed using various preprocessing methods including first deriva-tive, multiplication scatter correction, Savitzky-Golay first derivative algorithm, standard normal variate, smoothing, and mov-ing-window median. Standard deviation was used to select the optimal spectral region of 4 000-6 000 cm~(-1). Then principal com-ponent analysis was used for classification. Principal component analysis results showed that three types of samples could be dis-criminated completely and the accuracy almost achieved 100%. This study demonstrated that near infrared spectroscopy technol-ogy and chemometrics method could be a fast, efficient, and novel means to diagnose cancer. The proposed methods would be a promising and significant diagnosis technique of early stage cancer.