With the development of traditional Chinese medicine (TCM) syndrome knowledge accumulation and artificial intelligence (AI), this study proposes a holistic TCM syndrome differentiation model for the classification prediction of multiple TCM syndromes based on deep learning and accelerates the construction of modern foundational TCM equipment. We searched publicly available TCM guidelines and textbooks for expert knowledge and validated these sources using ten-fold cross-validation. Based on the BERT and CNN models, with the classification constraints from TCM holistic syndrome differentiation, the TCM-BERT-CNN model was constructed, which completes the end-to-end TCM holistic syndrome text classification task through symptom input and syndrome output. We assessed the performance of the model using precision, recall, and F1 scores as evaluation metrics. The TCM-BERT-CNN model had a higher precision (0.926), recall (0.9238), and F1 score (0.9247) than the BERT, TextCNN, LSTM RNN, and LSTM ATTENTION models and achieved superior results in model performance and predictive classification of most TCM syndromes. Symptom feature visualization demonstrated that the TCM-BERT-CNN model can effectively identify the correlation and characteristics of symptoms in different syndromes with a strong correlation, which conforms to the diagnostic characteristics of TCM syndromes. The TCM-BERT-CNN model proposed in this study is in accordance with the TCM diagnostic characteristics of holistic syndrome differentiation and can effectively complete diagnostic prediction tasks for various TCM syndromes. The results of this study provide new insights into the development of deep learning models for holistic syndrome differentiation in TCM.

Traditional Chinese medicine (TCM) has been widely used in the treatment of coronavirus disease 2019 (COVID-19). Tongue features should be referred for diagnosis and treatment of diseases in TCM. Therefore, it is necessary to analyze the tongue features of the patients suffered from COVID-19. COVID-19 Symptom Database (established by Evidence-based Medicine Center of Tianjin University of TCM) was searched for eligible tongue pictures. The tongue pictures were assessed by TCM experts to extract the data of tongue color, coating color, tongue body shape and coating proper feature. The relationship between tongue features and patient’s condition was analyzed. Patients diagnosed as mild and moderate COVID-19 commonly had light red tongue and white coating. Severe patients had purple tongue and yellow coating. The proportion of critical patients with tender tongue increased to 75%. Greasy coating was a significant characteristic of patients with COVID-19. The proportions of greasy coating were 53.33%, 73.30%, 83.67 and 87.5% in disease category of mild, moderate, severe and critical. In addition, the thick coating proportion increased from mild (24.89%) to critical (50.00%). Tongue features have certain relationship with the category of COVID-19. Tongue features can serve as potential indicators for the evaluation of patient’s condition and prognosis. Further studies are needed to enhance the quantification of tongue features and develop standards.

Tongue diagnosis is one of the four diagnostic methods of traditional Chinese medicine (TCM), which has important value in clinical disease diagnosis and efficacy evaluation. The change in tongue coating is a comprehensive cause of multi-dimensional changes such as color, texture, and substance. However, the color tongue image contains less spectral information, which may lead to the lack of key information in tongue diagnosis. Hyperspectral images can obtain reflection information of tongue images in hundreds of spectral bands. Unlike traditional color images, the rich spectral information can more accurately and sensitively describe and classify tongue coating, and has been widely applied in biomedical images. In this paper, we conducted feature extraction and analysis on hyperspectral images of different tongue coatings, and proposed a spectral-spatial feature deep learning framework to classification and quantitative recognition the tongue coating based on hyperspectral image features. Firstly, 360 hyperspectral images of tongue body were collected, and clinicians were identify all tongue coatings and divided them into 6 different grades. The hyperspectral features of each tongue coating area were extracted respectively. In order to reduce noise interference, singular spectrum analysis was used to preprocess the hyperspectral curve features. Considering the actual situation of tongue coating, a depth learning model was established to analyze the spectral and spatial feature of the hyperspectral tongue image to identified the grading of tongue coating. The experimental results showed that tongue coating with different quantization levels had different hyperspectral features, and the recognition rate of tongue coating quantization level can reach 87.21% using the spatial-spectral features.