Abstract:
OBJECTIVE To explore the construction of the physical identification model of twenty-five people of yin and yang by combining the hand diagnosis image analysis of machine learning, in order to enrich the input of physical identification features.
METHODS Based on the theory of twenty-five people of yin and yang, the hand images of 542 patients with cardiovascular disease were collected by self-developed hand diagnosis acquisition equipment. The key points of the hand were extracted by the Mediapipe palm key point algorithm, and the palm area was segmented by geometry to obtain the length and color gamut characteristics of different parts. The physician with clinical license was used to determined the patient's yin and yang twenty-five constitution according to the electronic medical record, and the patient's constitution was identified by comparing Random Forest, Logistic Regression, Xgboost algorithm and Light GBM.
RESULTS This study successfully constructed a TCM constitution identification model based on hand characteristics, which can identify ten basic constitution types including gold wood, gold fire, gold water, gold soil, wood fire, wood water, wood soil, water fire, water soil and fire soil. After sample equalization, the Random Forest model performed best in physical identification, with an accuracy rate of 0.69, which was significantly better than other models.
CONCLUSION The constitution identification model of TCM hand diagnosis based on hand characteristics can realize the constitution identification of the five elements of TCM. However, due to the use of only hand diagnosis data of TCM, the accuracy needs to be improved. Compared with the traditional body mass scale, it is more objective, convenient, highly stable and consistent. In the future, multimodal integration with face diagnosis and whole body inspection can be carried out to promote the accuracy, intelligence and multi-dimensional evaluation ability of TCM constitution identification technology.
下载: