基于粪便可见-近红外反射光谱的高山麝慢性肠炎诊断

提出了一种利用粪便可见-近红外反射光谱进行高山麝慢性肠炎诊断的新方法。以FieldSpec？3地物光谱仪采集了125份高山麝粪便（正常粪样70份，慢性肠炎患者粪样55份）的光谱数据，将其随机分成训练集（95份）和检验集（30份）。光谱经S.Golay平滑与一阶导数处理后以主成分分析法（PCA）降维。以前6个主成分（含原始光谱95.16%的特征信息）作为新变量，利用训练集样本，分别以模糊模式识别、BP-神经网络、Fisher线性判别以及Bayes逐步判别四种方法建立高山麝慢性肠炎的诊断模型。对检验集30个未知样的预测表明，Fisher线性判别的准确率为86.7%，模糊模式识别与BP-神经网络模型判别的准确率为90%，Bayes逐步判别的准确率最高，达93.3%。进一步分析发现所有误诊都源于将正常样误判为病样，四种方法对病样的检出率均达100%。说明利用粪便的可见-近红外反射光谱进行高山麝慢性肠炎的快速、非接触性诊断是可行的，且PCA 结合Bayes逐步判别是一种优选方法。

Diagnosis of Chronic Enteritis of Alpine Musk Deer(Moschus Chrysogaster) Based on Visible-Near Infrared Reflectance Spectra of Feces

A new method was put forward to diagnose chronic enteritis of alpine musk deer (Moschus chrysogaster) by visible-near infrared reflectance spectra of feces. A total of 125 feces samples, including 70 samples from healthy individuals (healthy samples) and 55 samples from chronic enteritis sufferers (diseased samples), were collected in Xinglongshan musk deer farm, Gansu province. The spectral scan was carried out in the darkroom (temperature 18 ℃-22 ℃, humidity 22%-25% and halogen lamp as a sole light source) with an ASD FieldSpec3 spectrometer. All the samples were divided randomly into two groups, one with 95 samples as the calibration set, and another with 30 samples as the validation set. The samples data were pretreated by the methods of S.Golay smoothing and first derivative. The pretreated spectra were analyzed by principal component analysis (PCA), and the top 6 principal components, which were computed by PCA and accounted for 95.16% variation of the original spectral information, were used for modeling as the new variables. The data of the calibration set were used to build models for diagnosing the chronic enteritis of alpine musk deer by means of back-propagation artificial neural network (ANN-BP), fuzzy pattern recognition, Fisher linear discriminant and Bayes stepwise discriminant, respectively. The predicted outcomes of the 30 unknown samples in validation set showed that the accuracy was 86.7% by the method of Fisher linear discriminant, 90% by fuzzy pattern recognition and ANN-BP model, and 93.3% by stepwise discrimination. Further analysis found that all misdiagnosed samples were derived from the healthy samples, which were treated as disease samples, and the detection rates of diseased samples were 100% by the four different methods. The results indicated that it was feasible to diagnose the chronic enteritis of alpine musk deer by visible-near infrared reflectance spectra of feces as a rapid and non-contact way, and the PCA combined with Bayes stepwise discriminant was a preferred method.