基于多级任务的“视觉脉诊”方法研究  

Methodological Study of Visual Pulse Diagnosis Based on Multi-Level Task

作　　者：赵智慧[1] 秦睿 李炜弘[3] 许强[1] 陈帅 王昕 王锭 Zhao Zhihui;Qin Rui;Li Weihong;Xu Qiang;Chen Shuai;Wang Xin;Wang Ding(Intelligent Medical College,Chengdu University of Traditional Chinese Medicine,Chengdu,610075,China;ZhenShu XiangCheng(Chengdu)Technology Company,Chengdu,610096,China;School of Basic Medical Science,Chengdu University of Traditional Chinese Medicine,Chengdu,6111137,China;Xin-Huangpu Joint Innovation Institute of Chinese Medicine,Guangzhou,510700,China)

机构地区：[1]成都中医药大学智能医学学院,四川成都611137 [2]真术相成(成都)科技有限公司,四川成都610096 [3]成都中医药大学基础医学院,四川成都611137 [4]广东省新黄埔中医药联合创新研究院,广东广州510700

出　　处：《成都中医药大学学报》2024年第5期60-66,共7页Journal of Chengdu University of Traditional Chinese Medicine

基　　金：国家自然科学基金项目(81873204);中国博士后科学基金面上项目“地区专项支持计划”(2022MD723720);四川省科技计划资助(2023NSFSC1813);成都市科技局技术创新研发项目(2021-YF05-01226-SN,2021-YF05-02035-SN);四川省中医药管理局科学技术研究专项课题(2021MS567);成都市医学科研课题(2020012)。

摘　　要：脉诊对中医诊断至关重要。然现有脉诊客观化高度依赖设备,阻碍其在远程在线诊疗中推广。受近年来基于远程光电容积脉搏波描记法(Remote Photoplethysmography,rPPG)的视频非接触式生理参数检测启发,本文提出了一种基于rPPG的“视觉脉诊”方法,以面部视频为输入预测中医脉象;研究通过多尺度融合及多任务学习进行ROI检测,利用深度卷积神经网络进行滤波获取纯净血容量脉搏波,后基于深度神经网络进行脉象预测。经真实病例验证后发现,基于深度学习及多级任务的“视觉脉诊”准确率可达85%以上,且优于其他rPPG法。“视觉脉诊”研究创新了非接触式脉诊研究范式,未来可推广至中医移动医疗,具有较大应用价值。Pulse diagnosis is crucial for Chinese medicine diagnosis.However,the existing objectivization of pulse diagnosis highly relies on equipment,which,unfortunately,hinders its promotion in mobile health.Inspired by the recent video-based non-contact physiological measurements based on remote photoplethysmography(rPPG),this paper proposes an rPPG-based "visual pulse diagnosis" method to predict the pulse with facial video.The study performed ROI detection through multi-scale fusion and multi-task learning,obtained rPPG-wave by filtering based on deep convolutional neural networks,and conducted pulse prediction using deep neural networks.After examination in real cases,we found that the accuracy of the deep learning and multilevel task-based "visual pulse diagnosis" can be over 85% and outperforms other rPPG methods."Visual Pulse Diagnosis" is an innovative paradigm for non-contact pulse diagnosis,which can be applied to mobile healthcare in Chinese medicine in the future with promising value.

关 键 词：视觉脉诊 多级任务 脉诊客观化 远程光电容积脉搏波描记法 

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]