摘要
为了对中医舌诊的客观化研究,提出了应用近红外光谱分析技术快速无创的对健康人、冠心病、糖尿病和肝炎患者的不同人群的舌诊近红外光谱进行识别的新方法。首先对98个样本光谱数据进行归一化处理,用主成分分析(PCA)方法得出的累计贡献率达99.88%的前8个主成分作为广义回归神经网络(GRNN)的输入变量,建立了舌诊近红外光谱的识别模型。利用该模型分别选取了18个不同人群的近红外光谱数据共72个样本用于神经网络的训练,余下的26个用于预测,当光滑因子为5/8时预测的最大误差为0.17342,最小误差为0,获得了较理想的预测精度。实验结果表明用PCA和GRNN相结合的方法对舌诊近红外光谱与疾病之间建立了较好的关联,对加强中医舌诊的客观化起到了很好的促进作用,为疾病的诊断提供了一种新的方法。
To make an objective research on tongue diagnosis of Tradition Chinese Medicine, it is proposed that spectroscopy analytical technique which is non-invasive discrimination tongue spectra of different groups who are healthy people and coronary, diabetic, hepatic patients. Firstly, normalize 98 samples of spectral data, next use principal component analysis to gain the first 8 principal components derived from cumulative contribution rate of 99.88%, then take the first 8 principal components as the general regression neural network input variables to establish a spectral discrimination of tongue model. It is selected according to the model that a total of 72 samples from 18 different groups of spectral data were used for neural network training, the remaining 26 used to predict. When the smooth factor is 5/8, the maximum error of prediction is 0.17342, the smallest error is 0, a more satisfactory prediction accuracy can be obtained. Experimental results show that the combination of PCA and GRNN, which reveals a good association between tongue diagnosis and disease spectrum, plays a good role in objectifying tongue diagnosis of Tradition Chinese Medicine, and provides a new approach for diagnosis.
出处
《红外技术》
CSCD
北大核心
2010年第8期487-490,494,共5页
Infrared Technology
基金
国家自然科学基金项目
编号:30973964
关键词
近红外光谱
舌诊
主成分分析
广义回归神经网络
near infrared spectroscopy, tongue diagnosis, principal component analysis, general regression neural network
