基于客观化多模态数据融合的亚健康肝郁脾虚证诊断模型构建  

Construction on diagnostic model for sub healthy liver depression and spleen deficiency syndrome based on objective multimodal data fusion

作　　者：丁艳 李阳[2] 王烨 白龙 冉清智 刘佳[1,3] 汪南玥 DING Yan;LI Yang;WANG Ye;BAI Long;RAN Qingzhi;LIU Jia;WANG Nanyue(Experimental Research Center,China Academy of Chinese Medical Sciences,Beijing 100700,China;Shaanxi University of Chinese Medicine,Xi’an 712046,China;Beijing Key Laboratory of TCM Basic Research on Prevention and Treatment of Major Diseases,Beijing 100700,China)

机构地区：[1]中国中医科学院医学实验中心,北京100700 [2]陕西中医药大学,西安712046 [3]中医药防治重大疾病基础研究北京市重点实验室,北京100700

出　　处：《中华中医药杂志》2025年第1期71-76,共6页China Journal of Traditional Chinese Medicine and Pharmacy

基　　金：中国中医科学院科技创新工程项目(No.CI2021A05207);北京市科技新星交叉合作课题(No.BJJC2022001);科技基础性工作任务书(No.2013FY114400);国家重点研发计划(No.2018YFC1707605);中央级公益性科研院所基本科研业务费专项资金项目(No.JBGS2021008)。

摘　　要：目的:基于客观化多模态数据融合构建亚健康肝郁脾虚证诊断模型。方法:以2023年3—12月在全国各地医疗机构及社区筛选亚健康志愿者共821例作为研究对象,分成肝郁脾虚证组与非肝郁脾虚证组。通过四诊采集设备,收集亚健康的四诊信息,建立亚健康中医客观化四诊数据库。数据处理后,在单模态分析上,脉诊采用主成分分析(PCA)、最小二乘回归(LS)与套索回归(Lasso)等方法进行判别;舌诊采用卷积神经网络进行分类;在多模态建模上,通过深度学习对亚健康肝郁脾虚证人群的脉、舌、声音等多源信息以全连接层与问诊进行拼接,从而实现诊断模型的构建。结果:在脉诊分析上,肝郁脾虚证组与非肝郁脾虚证组人群脉诊信息具有显著性差异,PCA差异主要在第2主成分上,LS及Lasso分类判别准确率在58%~62%之间;与非肝郁脾虚证组比较,肝郁脾虚证组的时域参数差异主要为双寸脉w减小,右关脉与左尺脉S、S2增加(P<0.05)。在舌诊分析上,肝郁脾虚证组的舌苔厚薄程度大于非肝郁脾虚证组(P<0.05)。此外,肝郁脾虚证单模态辨别准确率低于多模态组合;舌诊+脉诊+问诊三模态组合达到最高准确率为84.00%;舌诊+闻诊+脉诊+问诊的准确率为75.33%。结论:本研究可以较好地对亚健康肝郁脾虚证进行客观化诊断,为中医临床证候诊断模型研究提供了客观化的科学依据。Objective:To construct a diagnostic model for sub-health liver-qi stagnation and spleen deficiency syndrome based on objective multimodal data fusion.Methods:Between March and December 2023,we enrolled 821 sub-healthy volunteers from multiple healthcare institutions and community centers across China.The participants were divided into two groups:an intervention group with liver-qi stagnation and spleen deficiency syndrome and a control group without liver-qi stagnation and spleen deficiency syndrome.Four-diagnostic information was collected from sub-health subjects using diagnostic devices to establish an objective TCM database.For single-modal analysis,pulse diagnosis employed principal component analysis(PCA),least squares(LS),and Lasso regression for discrimination;tongue diagnoses utilized convolutional neural networks for classification.For multimodal modeling,deep learning was applied to integrate pulse,tongue,and voice information through fully connected layers with inquiry diagnosis.Results:In pulse analysis,significant differences were observed between groups,primarily in the second principal component of PCA,with LS and Lasso achieving 58% to 62% classification accuracy.The experimental group showed decreased w in both inch pulses and increased S and S2 in right guan and left chi pulses(P<0.05).Tongue coating thickness varied more significantly in the experimental group(P<0.05).Multimodal combinations outperformed single-modal diagnosis,with tongue+pulse+inquiry achieving the highest accuracy(84.00%),while the four-modal combination reached 75.33%.Conclusion:This study provides an objective diagnostic approach for sub-health liver-qi stagnation and spleen deficiency syndrome,offering scientific evidence for TCM syndrome diagnosis.

关 键 词：亚健康 肝郁脾虚证 四诊客观化 多模态数据融合 诊断模型 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程] R241[自动化与计算机技术—控制科学与工程]