基于长短时记忆网络及智能调优算法优化的冠心病患者中医脉象识别研究  

Study on pulse recognition of traditional Chinese medicine in patients with coronary heart disease based on long short-term memory and intelligent tuning algorithms

作　　者：吕仪 吴海妹 燕海霞[1,2] 洪静 侍煜景[4] 王忆勤 徐璡 LYU Yi;WU Haimei;YAN Haixia;HONG Jing;SHI Yujing;WANG Yiqin;XU Jin(School of Traditional Chinese Medicine,Shanghai University of Traditional Chinese Medicine,Shanghai 201203,China;Shanghai Key Laboratory of Health Identification and Assessment,Shanghai 201203,China;Shanghai Lingyun Sub-district Health Service Center,Shanghai 200237,China;Nanjing University of Chinese Medicine,Nanjing 210023,China)

机构地区：[1]上海中医药大学中医学院,上海201203 [2]上海市健康辨识与评估重点实验室,上海201203 [3]上海市徐汇区凌云街道社区卫生服务中心,上海200237 [4]南京中医药大学,南京210023

出　　处：《中华中医药杂志》2025年第1期77-82,共6页China Journal of Traditional Chinese Medicine and Pharmacy

基　　金：国家自然科学基金面上项目(No.81673880);上海市科委上海市健康辨识与评估重点实验室项目(No.21DZ2271000)。

摘　　要：目的:为早期检测冠心病探索无创、快速、经济、可靠的技术。方法:纳入健康人群196名,仅高血压病无其他心血管疾病患者186例,明确诊断冠心病患者226例,采集其脉象数据并提取参数。在前期研究基础上,对传统脉象参数进行扩充,引入更多特征并使用树模型特征选择对所有特征的重要性排序后取前十进行降维。随后基于长短时记忆网络(LSTM)算法以及使用麻雀搜索算法(SSA)、模拟退火算法(SA)、粒子群算法(PSO)以及开普勒优化算法(KOA)分别对其进行调优。结果:基于KOA优化的LSTM取得了0.83607的准确率,0.83446的精准率,0.83734的召回率,0.8359的F1得分以及0.91793的特异性。结论:基于KOA优化的LSTM相比其他基于LSTM的智能优化算法拥有更高的准确率、精准率、召回率、F1得分以及特异性,对健康人群、高血压病以及冠心病人群有较好的识别效果。Objective:To explore a non-invasive,rapid,economical,and reliable technique for the early detection of coronary heart disease(CHD).Methods:A total of 196 healthy individuals,186 patients with only hypertension and no other cardiovascular diseases,and 226 patients with a definite diagnosis of CHD were included.Pulse wave data was collected and parameters were extracted.Based on previous research,the traditional pulse wave parameters were expanded by introducing more features.Feature selection was performed using a tree-based model to rank the importance of all features,and the top ten features were selected for dimensionality reduction.Subsequently,based on the long short-term memory(LSTM) algorithm,and using sparrow search algorithm(SSA),simulated annealing(SA),particle swarm optimization(PSO),and Kepler optimization algorithm(KOA) for tuning,respectively.Results:The LSTM model optimized by KOA achieved an accuracy of 0.83607,precision of 0.83446,recall of 0.83734,F1-score of 0.8359,and specificity of 0.91793.Conclusion:Compared with other optimization algorithms,the LSTM model optimized by KOA has higher accuracy,precision,recall,F1-score,and specificity,demonstrating better recognition performance on healthy individuals,hypertension,and CHD patients.

关 键 词：开普勒优化算法 长短时记忆 冠心病 脉象识别 

分 类 号：R259[医药卫生—中西医结合]