舌体高光谱的高血压患者中合并冠心病的筛查方法

高血压合并冠心病是高血压疾病的一类严重合并症。目前高血压患者对合并冠心病的情况检测不及时，容易导致严重事件的发生，其致残率、致死率极高，严重影响患者的生活质量。因此对高血压患者合并冠心病的早期发现对预防更严重事件的发生显得尤为重要。舌体信息与心血管功能关系密切，为此，提出一种基于舌体高光谱用于高血压患者中合并冠心病筛查新方法。采集154例心血管科门诊患者舌体377.8～1 049.1 nm高光谱数据并记录临床诊断信息。对高血压患者组及高血压合并冠心病患者组舌体光谱比较发现，在500～600 nm波段之间，两组光谱存在显著性差异。基于中医舌诊分区理论，对舌体进行区域划分，划分为舌尖、舌左、舌中、舌右以及舌根五个区域。各舌体分区在波长509.6和561.2 nm以及540.5和576.7 nm均存在极大和极小值，这与血红蛋白对光的吸收特征相吻合。对组间各舌体区域光谱特征进行差异性分析。T检验分析结果显示高血压患者组与高血压合并冠心病患者组舌尖与舌中区域光谱具有显著性差异。基于高血压患者组与高血压合并冠心病患者组舌体不同区域光谱特征的差异性，利用神经网络，选取500～600 nm舌尖与舌中光谱数据进行后续高血压患者组与高血压合并冠心病患者组分类辨识模型的建立。选用准确度、灵敏度以及特异性作为模型预测能力的评价指标。最终获得模型的预测准确度、灵敏度以及特异性分别为84.78%，86.95%和82.61%。实验结果表明高血压患者与高血压合并冠心病患者舌体高光谱信息之间存在显著性差异。利用舌体高光谱所建立的分类模型能够用于高血压以及高血压合并冠心病患者的无创辨识，为高血压患者中高血压合并冠心病的早期发现提供一种有效筛查手段。

Study on Screening of Hypertension Complicated With Coronary Heart Disease Based on Tongue Hyperspectral Imaging

Coronary heart disease (CHD) is a serious complication of hypertension disease (HD). But this complication can not get timely detected, which is likely to cause serious events and result in extremely high disability rates and mortality. Therefore, early screening is important to provide early intervention to prevent the occurrence of serious events. Tongue manifestation can reflect the condition of the heart. In this paper, we applied tongue hyperspectral imaging to the early screen of the HD complicated with CHD patients. Hyperspectral data of 154 samples were collected from the department of cardiology. The wavelength range of hyperspectral data is 377.38~1 049.1 nm. The correlated clinical diagnosis was also recorded. Hyperspectral data of the tongue was separated into five parts according to the Chinese medicine theory, which included tongue tip, tongue left, tongue middle, tongue right and tongue root. There are maximum and minimum values at 509.6, 561.2, 540.5 and 576.7 nm. Moreover, this is consistent with the light absorption characteristics of hemoglobin.Then different tongue parts of the two groups were compared. The T-test results showed that there were significant differences between tongue tip and tongue middle in the wavelength region of 500~600 nm. Backpropagation artificial neural network (BPANN) was employed as an identification method for the screening whether or not complicated with CHD. The optimal results of screening model are obtained with an accuracy of 84.78%, sensitivity of 86.95%, and specificity of 82.61%, respectively. Experiment results showed that there were significant differences between HD and HD-CHD hyperspectral data, and hyperspectral imaging of the tongue provides a possible way for screening HD complicated with CHD among HD patients.