摘要
目的传统中医体质辨识主要依靠填写体质量表,存在一定的主观性,而后出现了基于舌象辨识体质的方法,为体质辨识客观化奠定了基石,但这些方法往往依赖于舌诊仪、面诊仪等专业设备。研究旨在提出一种仅依赖室内光学摄像头的基于人体骨骼特征的中医体质辨识模型,研讨将体质辨识的输入特征扩展至全身的可行性。方法采集受试者身高、体质量、性别、全身图像,并让受试者填写体质量表,以量表测试结果作为数据标注,利用深度学习算法提取人体骨骼特征,并利用神经网络建立身高、体质量、性别以及骨骼特征与受试者体质之间的联系,建立基于骨骼特征的中医体质辨识模型,并评估其表现。结果实验结果表明,基于人体骨骼特征的中医体质辨识模型在样本数据上取得了一定的准确性,在仅依靠室内光学摄像头的条件下,能够达到与基于舌象的体质辨识模型相近的水平。结论基于人体骨骼特征的中医体质辨识模型能够在仅依靠室内光学摄像头的条件下实现中医体质辨识,相较于传统的体质量表更加具有客观性,相较于基于舌象的方法更加便捷。未来可以对此类方法进行多模态融合,优势互补,进一步完善中医体质辨识客观化、智能化、多模态化。
Objective Traditional Chinese medicine(TCM) constitution identification mainly relies on filling out constitution questionnaires, which has a certain degree of subjectivity. Subsequently, the methods based on tongue and facial features for identifying constitutions have emerged, laying the foundation for objective constitution identification. However, these methods often rely on specialized equipment such as tongue and facial diagnosis instruments. This study aims to propose a TCM constitution identification model based on human skeletal features that only rely on an indoor optical camera and to explore the feasibility of extending the input features of constitution identification to the whole body. Methods This study collected the height, weight, gender and full-body posture images of the participants and asked them to fill out constitution questionnaires. The questionnaire results were used as data labels. Deep learning algorithms were employed to extract human skeletal features and neural networks to establish the relationship between height, weight, gender, skeletal features and the constitutions of. Then, it established a TCM constitution identification model based on skeletal features and evaluated its performance based on the constitution questionnaires filled out by the participants. Results The experimental results showed that the TCM constitution identification model based on human skeletal features achieved a certain degree of accuracy on the sample data and can achieve accuracy similar to that of the tongue-based constitution identification model with only an indoor optical camera. Conclusion The TCM constitution identification model based on human skeletal features can achieve constitution identification with only an indoor optical camera and is more objective than traditional constitution questionnaires and more convenient than methods based on tongue and facial features. In the future, multimodal fusion of such methods can complement each other's strengths and further improve the objectivity, intelligence and multimodality of TCM constitution identification.
作者
王子琰
李红岩
周作建
管树桃
郎许锋
杨涛
胡孔法
WANG Ziyan;LI Hongyan;ZHOU Zuojian;GUAN Shutao;LANG Xufeng;YANG Tao;HU Kongfa(College of Artificial Intelligence and Information Technology,Nanjing University of Chinese MedicineNanjing 210023,Jiangsu,China;Institute of Chinese Medicine Literature,Nanjing University of Chinese Medicine,Nanjing 210023,Jiangsu,China;Jiangsu Collaborative Innovation Center of Traditional Chinese Medicine in Prevention and Treatment of Tumor,Nanjing 210023,Jiangsu,China)
出处
《中华中医药学刊》
CAS
北大核心
2024年第8期37-40,I0008,I0009,共6页
Chinese Archives of Traditional Chinese Medicine
基金
国家重点研发计划重点专项(2022YFC3502302)
江苏省研究生科研创新计划项目(KYCX23_2073)。
关键词
中医体质辨识
人体骨骼特征
深度学习
姿态估计
TCM constitution identification
human skeletal features
deep learning
pose estimation
