【摘要】 目的:提高肺癌患者和健康人血清表面增强拉曼光谱(SERS)分类模型的稳定性和分类精度。方法:利用递归特征选择、连续投影算法、竞争性自适应重加权算法与主成分分析对血清SERS光谱进行特征波长提取，并结合深度神经网络、偏最小二乘判别分析以及支持向量机算法建立分类模型。结果:递归特征选择以及竞争性自适应重加权算法提高分类模型的稳定性的效果较为明显，对于竞争性自适应重加权算法，深度神经网络分类模型的训练集交叉验证准确率为94.55%,测试集准确率为93.75%,敏感性为87.5%,特异性为100%,优于其余两种模型。结论:通过特征波长提取方法建立分类模型实现了对肺癌患者与健康人血清SERS光谱的有效识别。更多还原

【Abstract】 Aims: This paper aims to improve the stability and classification accuracy of serum surface enhanced Raman spectroscopy(SERS) classification models for lung cancer patients and healthy individuals. Methods: Recursive feature selection, the successive projections algorithm, the competitive adaptive reweighted sampling algorithm and principal component analysis were used to select the spectral features. Then the deep neural network, partial least squares discriminant analysis, and the support vector machine classification algorithm were used to establish a Raman spectral classification model for lung cancer serum. Results: Recursive feature selection and the competitive adaptive reweighting algorithm had significant effects on improving the stability of the classification models. For the competitive adaptive reweighting algorithm, the training set cross-validation accuracy of the deep neural network classification models was 94.55%. The test set accuracy was 93.75%. The sensitivity was 87.5%; and the specificity was 100%, which was superior to the other two models. Conclusions: A classification model based on feature wavelength extraction was established to effectively identify SERS spectra from lung cancer patients and healthy individuals.更多还原