乳腺癌患者血清的傅里叶变换中红外光谱研究

目的　基于傅里叶变换中红外光谱技术建立一种有效区分和鉴别健康人群和乳腺癌患者的方法。方法　采集86例健康女性和85例乳腺癌患者的血清样品光谱,绘制两类人群血清样品光谱图;对两类血清样品分别提取主成分信息,绘制主成分得分2D和3D散点图,进一步计算主成分1~10的得分（PC1~PC10）;利用判别分析法原理建立判别分析模型,基于马氏距离对所有样本进行判别;在波数范围3 931~619 cm-1内计算不同光谱预处理方式下所建模型的性能指标评分,选择最佳预处理方式建立模型。结果　女性健康人群和女性乳腺癌患者的光谱在波数 3 363 cm-1、2 360 cm-1、1 641 cm-1、1 552 cm-1、663 cm-1处的峰强差异均有统计学意义（P＜0.05）;主成分分析结果显示,2组人群PC1~PC4差异有统计学意义（P＜0.05）,PC5~PC10差异无统计学意义（P＞0.05）;与正常组相比,乳腺癌组患者到N的马氏距离值高,到C的马氏距离值低（P＜0.05）;所建模型的验证集正判率为100%;对光谱不进行任何处理下所建模型最优,性能指标评分为94.1分。结论　傅里叶变换中红外光谱法可用于区分和鉴别健康人群和乳腺癌患者,有望成为乳腺癌辅助诊断的一种方法。

Study on Fourier transform mid-infrared spectroscopy in serum of breast cancer patients

Objective　To establish an effective method for distinguishing and identifying normal people and breast cancer patients based on Fourier transform mid-infrared spectroscopy technology. Methods　The serum samples of 86 female normal people and 85 female breast cancer patients were collected, and the spectra of the serum samples were drawn for the two groups of people. The principal component information of the two types of serum samples were extracted. The principal component score 2D and 3D scatter plots were drawn, and the scores of principal component 1-10 (PC1-PC10) were further calculated. Using the principle of discriminant analysis to establish a discriminant analysis model, and all samples were judged based on Mahalanobis distance. The performance index scores of models built under different spectrum preprocessing methods were calculated within the wave number range of 3 931-619 cm-1, and the best preprocessing method was selected to build the model. Results　There were significant differences in the peak intensities of the spectrum at wave numbers 3 363 cm-1, 2 360 cm-1, 1 641 cm-1, 1 552 cm-1, and 663 cm-1 between female normal population and female breast cancer patients (P＜0.05). The results of principal component analysis showed that there were significant differences in PC1-PC4 between two groups (P＜0.05), and there were no significant differences in PC5~PC10 between two groups (P>0.05). Compared with the normal group, the Mahalanobis distance to N in the breast cancer group is higher, and the Mahalanobis distance to C is lower (P＜0.05), and the positive judgment rate of the validation set of the built model was 100%. The model built without any processing on the spectrum was the best, and the performance index score was 94.1 points. Conclusion　Fourier transform mid-infrared spectroscopy can be used to identify and distinguish normal people and breast cancer patients, and it is expected to become a method for assisting breast cancer diagnosis.