基于高光谱图像的中医舌色分级量化研究  

Study on the Quantification of Traditional Chinese Medicine Tongue Color Classification Based onHyperspectral Images

作　　者：张冬 庞稳泰 王可仪 杨丰文 张俊华 ZHANG Dong;PANG Wentai;WANG Keyi;YANG Fengwen;ZHANG Junhua(Center for Evidence-Based Medicine,Tianjin University of Traditional Chinese Medicine,Tianjin 300193,China;Xin-Huangpu Joint Innovation Institute of Chinese Medicine,Guangzhou 510000,China)

机构地区：[1]天津中医药大学循证医学中心,天津300193 [2]广东省新黄埔中医药联合创新研究院,广州510000

出　　处：《世界科学技术-中医药现代化》2024年第7期1925-1930,共6页Modernization of Traditional Chinese Medicine and Materia Medica-World Science and Technology

基　　金：广东省新黄埔中医药联合创新研究院联合创新研究项目(2022IR024):基于舌参数分析的中医药疗效评价指标与测量设备研制,负责人:张俊华;国家自然科学基金委员会青年科学基金项目(82205316):基于高光谱图像和深度学习的中医舌诊参数量化研究,负责人:张冬。

摘　　要：目的中医舌诊在疾病临床诊疗中具有重要的作用,但是当前研究结果并不适用于临床疗效评价中。本研究基于高光谱图像技术对舌色进行了特征提取和颜色等级量化研究,使其适用于舌色量化描述。方法制定纳入排除标准,获取400-1000 nm的128个不同光谱波长的舌象高光谱图像,通过中医临床专家对红舌与黄苔进行4种不同颜色程度(轻度、中度、重度、严重)的判别,最后建立基于机器学习模型的红舌与黄苔等级量化预测模型。结果不同颜色程度等级的红舌与黄苔在高光谱曲线特征上存在显著的差异,可以作为颜色等级量化的基础,借助主成分分析+随机森林模型能够实现85.79%和88.34%的红舌与黄苔不同颜色等级的预测。结论借助高光谱图像数据特征与机器学习模型进行舌色的不同颜色等级预测,取得了较好的准确性。Objective Traditional Chinese medicine tongue diagnosis plays an important role in the clinical diagnosis and treatment of diseases,but the current research results are not applicable to the evaluation of clinical efficacy.This study conducted a hierarchical quantitative study on tongue color based on hyperspectral data of tongue images,making it suitable for clinical efficacy evaluation.Methods Establish inclusion and exclusion criteria,obtain tongue images of different spectral wavelengths within the visible light range of 400-1000 nm,and use traditional Chinese medicine clinical experts to distinguish between red tongue and yellow coating in four different color levels(mild,moderate,severe,and severe).Finally,establish a quantitative prediction model for the grade of red tongue and yellow coating based on machine learning models.Results There were significant differences in hyperspectral curve characteristics between red tongue and yellow coating with different color levels,which could be used as the basis for grade quantification.With the help of principal component analysis+random forest model,85.79%and 88.34%of the red tongue and yellow coating with different color levels could be predicted.Conclusion The use of hyperspectral image data features and machine learning models for predicting different color levels of tongue color has achieved good accuracy.

关 键 词：中医舌诊 高光谱特征 红舌黄苔 指标量化 特征提取 

分 类 号：R241.25[医药卫生—中医诊断学]