Development of a cold-heat syndrome classification model for children with allergic rhinitis based on multimodal data

Introduction

Allergic rhinitis (AR) is a common chronic disease in childhood. Globally, the prevalence of AR in children ranges from 2% to 25%. The prevalence of AR in children is rising, and younger children are being affected. In China, the prevalence of AR in children is estimated at 18.46%, reaching 22.77% in some regions (1-4).

AR can significantly impact a child’s quality of life, and if untreated, AR can lead to serious complications such as bronchial asthma and obstructive sleep apnea, which pose a risk of mortality (5,6). Worldwide, AR in children is a clinical and economic burden for affected families and society. In the United States, the direct and indirect costs of treating AR in children are estimated at $24.8 billion per year. In Korea, the National Health Insurance Agency showed the annual medical cost of AR treatment for children under 18 years increased from $50 million to $131.7 million over 10 years (7,8).

There remains an unmet clinical need for efficacious and cost-effective treatments for AR in children to reduce the global burden of the disease. Although current treatment using Western medicines can alleviate symptoms, the recurrence rate is high, and some drugs have side effects in children (9,10). Traditional Chinese medicine (TCM) is an effective and inexpensive treatment for AR in children. A large number of studies have shown that TCM can effectively prevent and treat AR in children, and significantly reduce its recurrence (11-15).

TCM has been widely used in the treatment of AR in children, with a particular emphasis on syndrome differentiation and treatment (16-18), which enables TCM practitioners to tailor treatments to the individual needs of a patient, potentially leading to better outcomes. In TCM theory, AR is classified under the category of “Bi Qiu” (nasal congestion and discharge), with its pathogenesis often involving dysfunctions of the lung, spleen, and kidney, as well as imbalances between cold and heat (19). Due to the unique physiological characteristics of children, disease progression tends to be rapid following onset. Therefore, timely and accurate differentiation between cold and heat syndromes is a critical step in determining the nature of the nasal condition and implementing effective TCM-based treatment strategies. TCM syndromes are differentiated using four main diagnostic methods (observation, auscultation-olfaction, interrogation, and pulse-palpation), with observation and interrogation particularly important in pediatric diagnosis. Observation of the tongue is especially crucial as it is a simple and intuitive way to obtain valuable clinical information. It is an important and frequently used non-invasive technique that can evaluate the condition of a patient’s internal organs. By analyzing the tongue’s color, coating, shape, and other characteristics, TCM practitioners can gain insight into the patient’s overall health and decide an appropriate treatment strategy (20-27).

Clinical application of visual inspection of the tongue is limited due to the lack of objective quantitative evidence (25,28); therefore, several studies have applied machine learning methods and artificial intelligence (AI) to enhance tongue diagnosis. One study developed an intelligent tongue diagnosis system based on the Cv-Swin Transformer architecture. By integrating TCM diagnostic methods with modern AI technology, tongue images were classified into ten categories according to TCM diagnostic criteria. The final model achieved an average accuracy rate of 87.37% in tongue diagnosis classification, demonstrating significantly better performance compared to conventional models (29). Another study developed heterogeneous integrated learning models using five machine learning algorithms as foundational classifiers, ultimately establishing an effective TCM constitution recognition model through objective tongue image features and machine learning techniques (30). A third study applied large language models (LLMs) in tongue diagnosis, providing a contextual multi-task learning approach. A structured prompt-label framework was constructed based on a tongue coating color analysis dataset and a small-sample constitution-labeled dataset, which could be used for complex TCM diagnostic tasks (31).

Currently, syndrome differentiation in TCM relies heavily on clinical experience and subjective judgment, which often leads to inconsistencies in diagnostic outcomes among different practitioners (32). This subjectivity hampers the standardization and objectification of TCM and limits its broader application in modern clinical practice (33-35). To achieve intelligent assistance and precision in TCM syndrome differentiation, it is imperative to integrate computer vision with big data technologies, thereby enhancing the scientific validity and reproducibility of diagnostic processes (28,36,37). A necessary prerequisite to the treatment of AR in children is to accurately classify them as cold syndrome or heat syndrome, which is an important element in TCM theory. Compared with approaches that rely solely on tongue images or clinical inquiry data, a multimodal model incorporating multidimensional information is more consistent with the TCM diagnostic principle of “inspection, listening/smelling, inquiry, and palpation” (38). Accordingly, this study aims to develop an intelligent syndrome differentiation model that integrates tongue image analysis, clinical symptoms, and medical history. The proposed model is designed to assist clinicians in accurately distinguishing between cold and heat syndromes in pediatric patients with AR. We present this article in accordance with the TRIPOD reporting checklist (available at https://tp.amegroups.com/article/view/10.21037/tp-2025-397/rc).

Methods

Participants

A total of 485 patients who visited the outpatient clinic of the Children’s Hospital, Zhejiang University School of Medicine between February 2023 and June 2025 and were diagnosed with AR based on the “Guidelines for the Diagnosis and Treatment of Allergic Rhinitis in Children (2022, revised edition)” and the “Traditional Chinese Medicine Diagnosis and Treatment Guidelines for Rhinitis in Children [2023]” (39,40) were eligible for this study. Inclusion criteria were: (I) age ≤18 years; (II) a diagnosis of AR in accordance with the aforementioned guidelines; and (III) written informed consent obtained from the patients’ parent or legal guardian for the patient’s voluntary participation. Exclusion criteria included serious medical conditions such as malignancy, metabolic liver disease, and/or autoimmune liver disease, participation in other clinical studies, or missing or incomplete clinical or imaging data. Finally, 391 patients were included in the analysis. Three TCM practitioners, each from tertiary Grade A hospitals in China with over 10 years of clinical experience, classified the patients into cold syndrome (n=92) and heat syndrome (n=299) according to the “Traditional Chinese Medicine Diagnosis and Treatment Guidelines for Rhinitis in Children [2023]” (40). Data collected from February 2023 to October 2023 were used as the model development set, which included 46 cases of cold syndrome and 130 cases of heat syndrome. Data from November 2023 to June 2025 constituted the model validation set, consisting of 46 cases of cold syndrome and 169 cases of heat syndrome. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The study was approved by the Institutional Review Board of the Children’s Hospital, Zhejiang University School of Medicine (No. 2022-IRBAL-219). Informed consent was obtained from all of the patients’ parents or legal guardians.

Data collection

Patients provided tongue images and completed a survey in a quiet room in the clinic. Tongue images were acquired after a 30-min fast with the same device in the same room and under natural light for all patients. The survey collected patient demographic and clinical information, patient and family medical history, and included a visual analog scale (VAS), the rhinoconjunctivitis quality of life questionnaire (RQLQ), which assesses symptoms in children with AR, as recommended by the “Guidelines for the diagnosis and treatment of AR in children (2022, revised edition)” (39), and a TCM syndrome questionnaire, which classifies children with AR as heat syndrome (heat syndrome in the lung meridian) and cold syndrome (wind-cold syndrome in the lung meridian, lung-spleen deficiency syndrome, and lung-kidney yang deficiency syndrome) according to the “Traditional Chinese Medicine Diagnosis and Treatment Guidelines for Rhinitis in Children [2023]” (40) (Table 1).

Image acquisition and preprocessing

Tongue images underwent preprocessing to adjust for variations in ambient lighting conditions and accommodate instances where patients did not fully protrude their tongues (Figure 1). First, Adobe Fireworks was used to manually segment the tongue area in the image. Then, the minimum enclosing rectangle covering the segmented area was drawn. The original image was cropped, resized to 512×512 pixel files, scaled to [0,1], and the pixels of the three channels of the red, green, and blue (RGB) image were normalized using mean = [0.485, 0.456, 0.406] and standard deviation = [0.229, 0.224, 0.225].

Figure 1 Image preprocessing (This image is published with the patient/participant’s parent’s consent).

Model development

The classification model was developed using a transformer-based model to integrate features extracted from the tongue images and information abstracted from the surveys (Figure 2). Tongue image features were identified with hybrid Dense Convolutional Network model with a Squeeze-and-Excitation (SE-DenseNet) module (base network, DenseNet121; dropout rate, 0.5; growth rate, 4; training epochs, 1,000; batch size, 20; learning rate, 0.1; learning rate updated every 500 training iterations) and converted to feature vectors. Patient demographic and clinical information and patient and family medical history were screened with the independent sample t-test, features with P<0.05 were selected and projected into a new vector space through a fully connected layer (FC Layer). Next, projected non-image features were prepended as a special token at the beginning of the sequence and concatenated with the feature sequence extracted from the images to form a complete input sequence. Finally, classification was performed using the Transformer model.

Figure 2 Model framework. (A) Classification model based on multi-modal data; (B) SE DenseNet. FC Layer, fully connected layer; SE-DenseNet, Squeeze-and-Excitation-DenseNet.

To facilitate a robust evaluation of the performance of our proposed multimodal model that included non-image and image-based features, two baseline models were also assessed: a multivariate Cox regression model utilizing non-image features and a deep learning model employing SE-DenseNet architecture for image-based analysis.

Model evaluation and visualization

Performance of the classification model was evaluated by calculating the area under the receiver operating characteristic (ROC) curve (AUC), as well as accuracy, precision, recall and F1 score, according to the following formulae, where true positive (TP), true negative (TN), false positive (FP) and false negative (FN) represent true positive, true negative, false positive and false negative, respectively.

Accuracy=TP+TNTP+FP+TN+FN[1]

Precision=TPTP+FP[2]

Recall=TPTP+FN[3]

F1=2∗Precision∗RecallPrecision+Recall[4]

A confusion matrix was employed to visually represent the classification outcomes of the model.

Computational resources

This study used a server in a Linux environment (Ubuntu 7.5.0). The server configuration was CPU Intel 4215FR 3.20 GHz, 64GB DDR4 memory, and the graphics card was 8 RTX 3060 Ti.

Statistical analysis

Python 3 version 3.6.13 (Python Software Foundation) was used as the programming language. PyTorch (https://pytorch.org/) was used for deep learning. The main package included python-opencv (version 1.0.0.14), scikit-learn module version 0.20.4, torch module version 1.10.1, and torchvision module version 0.11.2. Continuous variables in the demographic and clinical dataset were screened for significant differences between patients with cold or heat syndrome using the independent sample t-test. Features with a P value of less than 0.05 were considered statistically significant and were selected for subsequent model development. This univariate feature selection process aimed to identify the most discriminative clinical variables for syndrome classification.

Results

Patient information

A total of 32 features were extracted based on patients’ demographic and clinical data and patient and family medical history using the independent sample t-test, These included clear runny nose (P<0.001), thick runny nose (P<0.001), fear of heat (P=0.001), fear of cold (relieved by adding more clothes and quilts, internal injury, internal syndrome, yang deficiency syndrome) (P<0.001), spontaneous sweating (frequent sweating when awake, especially during activities, often seen in qi deficiency and yang deficiency) (P<0.001), thirst for cold water (P=0.04), abdominal pain (P=0.001), abdominal distension (P<0.001), poor appetite (P=0.03), dry and hard stool, constipation (P=0.001), thin sputum (P<0.001), thick sputum (P=0.02), dry mouth and fever (P<0.001), nose bleeding (P=0.044), fear of wind and cold (P<0.001), shortness of breath and laziness (P<0.001), low and timid voice (P=0.005), weight loss (P<0.0001), fatigue and weakness in limbs (P=0.003), red nasal mucosa (P<0.001), light red nasal mucosa (P<0.001), watery nasal discharge (P<0.001), red throat (P<0.001), pale complexion (P=0.02), sallow complexion (P=0.003), red tongue (P<0.001), pale or light red tongue (P<0.001), pale and fat tongue (P<0.001), thin white fur (P<0.001), yellow fur (P=0.02), rapid pulse (P<0.001), and weak pulse (P<0.001).

Model evaluation

Figure 3 presents the ROC curves of three models across both the training and test sets, illustrating the model’s discriminative ability. The detailed performance metrics for each model are summarized in Table 2. Specifically, the Non-Image Model, which used only clinical features, achieved AUCs of 0.805 and 0.792 in the training and test sets, respectively, with corresponding accuracy, precision, recall, and F1 scores of 0.790, 0.931, 0.391, 0.867 (training set) and 0.814, 0.899, 0.500, 0.884 (test set). In contrast, the Image Model, which relied on image-based features, showed improved performance, with AUCs of 0.843 (training) and 0.837 (test), and accuracy, precision, recall, and F1 scores of 0.818, 0.866, 0.757, 0.885 (training set) and 0.828, 0.876, 0.652, 0.889 (test set), respectively.

Figure 3 ROC of three models. (A) Training set; (B) test set. AUC, area under the curve; ROC, receiver operating characteristic.

Our multimodal model, which integrates both clinical and image-based features, outperformed both single-modality models. In the training set, the multimodal model achieved an AUC of 0.931, accuracy of 0.875, precision of 0.949, recall of 0.869, and F1 score of 0.920. The test set results were similar, with an AUC of 0.877, accuracy of 0.856, precision of 0.863, recall of 0.829, and F1 score of 0.910. These results indicate the superior ability of the multimodal model to leverage complementary information from both clinical and imaging features.

Figure 4 displays the confusion matrices for each model, highlighting the improvements in classification accuracy and the reduction in misclassifications, particularly for the test set. The proposed multimodal approach, which incorporates both clinical and image-based data, not only better simulates the TCM diagnostic process but also demonstrates robust performance across different evaluation metrics, underscoring its potential clinical applicability.

Figure 4 Confusion matrix. (A) Multimodal model (training set). (B) Multimodal model (test set). (C) Non-image model (training set). (D) Non-image model (test set). (E) Image model (training set). (F) Image model (test set).

Discussion

This study established a model based on multimodal data that classifies children with AR as cold syndrome or heat syndrome. The model used features extracted from tongue images, patient demographic and clinical information, and patient and family medical history. The model demonstrated excellent performance in the training set and test set (AUCs of 0.931 and 0.877, respectively). We constructed three distinct models based on clinical data, tongue image data, and a multimodal approach, respectively. The final results indicated that the multimodal model achieved the best performance, which aligns with the TCM diagnostic approach of integrating inspection, listening and smelling, inquiry, and palpation. The multimodal model established in this study outperformed others when classifying children with AR as cold syndrome or heat syndrome, as it had the best AUC, accuracy, precision, recall and F1 score.

Cold-heat differentiation is not merely a classification of individual symptoms, but rather a concrete embodiment of the holistic view and the core principle of syndrome differentiation and treatment with TCM theory (41), “Cold and heat syndrome differentiation” aligns with the basic theory of TCM, which categorizes the causes of diseases into external causes and internal causes and differentiates syndromes according to eight principles [yin, yang, exterior, interior, cold, heat, deficiency (xu) and excess (shi)] (42). According to the Guidelines for the Diagnosis and Treatment of Allergic Rhinitis in Children (Revised) 2023 (40), AR in children can be divided into “wind-cold syndrome of lung meridian”, “latent heat syndrome of lung meridian”, “deficiency of lung and spleen qi” and “deficiency of lung and kidney yang”. The basis for distinguishing these four syndromes is to differentiate between cold syndrome and heat syndrome. Among them, “latent heat syndrome of lung meridian” represents heat syndrome, and “wind-cold syndrome of lung meridian”, “deficiency of lung and spleen qi” and “deficiency of lung and kidney yang syndrome” represent cold syndrome. In clinical practice, cold syndrome is relatively rare in children, consistent with TCM theory that children have a “pure yang body” (43). In clinical practice, cold and heat syndromes should be used very cautiously. One should not be overly confined by the notion of “children being purely yang in nature”. For example, if a pediatrician observes that a child has had AR for a prolonged period, signs of lung qi deficiency or spleen and kidney yang deficiency may emerge, making the protection of yang qi particularly important (44). Another clinical study involving 84 children with AR and lung qi deficiency with cold syndrome showed that the “Traditional Chinese Medicine Nasal Disease Sequential Therapy” combined with acupuncture significantly improved primary symptoms and quality of life (P<0.05) compared with western medication, demonstrating superior therapeutic efficacy (45).

When exogenous cold pathogens invade a child’s body, they may suffer from wind-cold colds, with symptoms of cold, no sweating, and headaches. When exogenous heat pathogens invade a child’s body, they may suffer from wind-heat colds, with symptoms of heat, such as fever, sore throat, and thirst. Identifying children with cold or heat syndrome allows preliminary determination of the cause of a disease and provides a decisive direction for subsequent treatment. TCM treats AR in children through syndrome differentiation, such that “heat should be treated with cold, and cold should be treated with heat”. Incorrect identification of cold and heat syndromes will result in prescribing faults and prescription errors. Warm and hot drugs are often used to treat cold syndrome. Misuse of cold and cool methods will worsen the condition. Heat-clearing methods are often used to treat heat syndrome. Misuse of warm and hot methods will cause an excess heat pattern. Accurately distinguishing cold and heat syndromes in children with AR can help healthcare providers choose appropriate treatment methods to achieve the best therapeutic effect.

Children are in a stage of growth and development. They have a “soft body build” and “weak body resistance” (46). Misdiagnosis of AR can seriously affect the quality of life of children. Improper treatment could lead to serious complications, such as bronchial asthma and obstructive sleep apnea. In severe cases, there is a risk of death <Zhao, 2023 #15533}. Therefore, early identification of cold and heat syndromes in children with AR can allow timely elimination of pathogens, prevent the spread of disease, and facilitate children’s recovery.

Currently, there is no auxiliary diagnostic method specifically for the cold-heat pattern differentiation in children. Cold-heat differentiation, as a core component of TCM syndrome differentiation, reflects the holistic and individualized nature of TCM treatment. The integration of multimodal data fusion technology is a modern interpretation of this traditional theory. By combining tongue diagnosis, clinical symptoms, medical history, and other multidimensional data, it enables more precise differentiation of cold-heat patterns, reducing subjective judgment errors and enhancing the efficiency and consistency of clinical diagnosis. This TCM-assisted diagnostic model, based on modern technology, not only inherits the essence of TCM syndrome differentiation but also promotes the objective and standardized development of TCM, providing new possibilities for the internationalization and modernization of TCM.

Limitations of the study

This study had several limitations. First, this study was conducted at a single center with small sample size, which may limit the generalizability of the model. Future studies should be conducted in multiple centers with larger sample sizes to confirm our findings. Second, the small sample size of the training set may have led to overfitting. Third, the data included in this study are all retrospective, which may have led to information bias. Future research needs to be conducted clinically and validated and optimized through prospective data.

Conclusions

This study introduces a multimodal decision-support model that integrates TCM syndrome differentiation with machine learning techniques. The model was developed in three stages. First, SE-DenseNet was used to extract features from tongue images. Second, the independent sample t-test was used to screen and select relevant features from patient demographic and clinical information and patient and family medical history. Third, a transformer model was used to integrate the features for cold and heat syndrome classification. By combining digital tongue image analysis with clinical data, the model can accurately distinguish cold or heat syndromes in pediatric AR. The model’s high accuracy demonstrates its potential to enhance the objectivity and efficiency of TCM diagnostics, offering a promising tool for modernizing TCM practices in clinical settings.

Acknowledgments

None.

Footnote

Reporting Checklist: The authors have completed the TRIPOD reporting checklist. Available at https://tp.amegroups.com/article/view/10.21037/tp-2025-397/rc

Data Sharing Statement: Available at https://tp.amegroups.com/article/view/10.21037/tp-2025-397/dss

Peer Review File: Available at https://tp.amegroups.com/article/view/10.21037/tp-2025-397/prf

Funding: This study was supported by the Key R&D Program of Zhejiang (Nos. 2023C03101 and 2023C03042), the Key Discipline Construction Project of Traditional Chinese Medicine by the Zhejiang Provincial Health Commission (grant No. 2024-XK-48), the Zhejiang Provincial Administration of Traditional Chinese Medicine Provincial Co-construction Project-Key Project (grant No. GZY-KJS-ZJ-2025-042), and the Key Research Project of Zhejiang Provincial TCM Science and Technology Plan (grant No. 2023ZF123).

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://tp.amegroups.com/article/view/10.21037/tp-2025-397/coif). The authors have no conflicts of interest to declare.

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The study was approved by the Institutional Review Board of the Children’s Hospital, Zhejiang University School of Medicine (No. 2022-IRBAL-219) and informed consent was obtained from all of the patients’ parents or legal guardians.

Open Access Statement: This is an Open Access article distributed in accordance with the Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International License (CC BY-NC-ND 4.0), which permits the non-commercial replication and distribution of the article with the strict proviso that no changes or edits are made and the original work is properly cited (including links to both the formal publication through the relevant DOI and the license). See: https://creativecommons.org/licenses/by-nc-nd/4.0/.

References

1. Wang Z. Interpretation of the treatment part of Clinical Practice Guideline: Diagnosis and Treatment in Children with Allergic Rhinitis. Chinese Journal of Practical Pediatrics 2019;34:169-75.
2. Wang RK, Liang JQ, Han W, et al. Zhonghua Yu Fang Yi Xue Za Zhi 2022;56:784-93. [Crossref] [PubMed]
3. Bousquet J, Schünemann HJ, Togias A, et al. Next-generation Allergic Rhinitis and Its Impact on Asthma (ARIA) guidelines for allergic rhinitis based on Grading of Recommendations Assessment, Development and Evaluation (GRADE) and real-world evidence. J Allergy Clin Immunol 2020;145:70-80.e3. [Crossref] [PubMed]
4. Schuler Iv CF, Montejo JM. Allergic Rhinitis in Children and Adolescents. Immunol Allergy Clin North Am 2021;41:613-25. [Crossref] [PubMed]
5. Zhao X, Zhang J, Qin Y-h, et al. Guideline for the Diagnosis and Treatment of Allergic Rhinitis in Children with Integrated Traditional Chinese and Western Medicine. Journal of Nanjing University of traditional Chinese Medicine 2023;39:274-84.
6. D'Elia C, Gozal D, Bruni O, et al. Allergic rhinitis and sleep disorders in children - coexistence and reciprocal interactions. J Pediatr (Rio J) 2022;98:444-54. [Crossref] [PubMed]
7. Brown T. Diagnosis and Management of Allergic Rhinitis in Children. Pediatr Ann 2019;48:e485-8. [Crossref] [PubMed]
8. Kim DH, Park YS, Jang HJ, et al. Prevalence and allergen of allergic rhinitis in Korean children. Am J Rhinol Allergy 2016;30:72-8. [Crossref] [PubMed]
9. Feng L, Lin L, Wang S, et al. Clinical practice guidelines for the treatment of allergic rhinitis in children with traditional Chinese medicine. Anat Rec (Hoboken) 2021;304:2592-604. [Crossref] [PubMed]
10. Tosca MA, Trincianti C, Naso M, et al. Treatment of Allergic Rhinitis in Clinical Practice. Curr Pediatr Rev 2024;20:271-7. [Crossref] [PubMed]
11. Carnazza M, Werner R, Begley M, et al. Traditional Chinese Medicine for Pediatric Allergic Diseases. Curr Allergy Asthma Rep 2025;25:33. [Crossref] [PubMed]
12. Dou YZ, Dai QG. Zhongguo Zhong Yao Za Zhi 2025;50:1419-26. [Treatment of allergic rhinitis in children by attenuating oxidative stress from sweat pore-Qi-triple energizer system].
13. Luo C, Wang Y, He B, et al. Exploring the Core Prescription and Underlying Mechanism of Traditional Chinese Medicine in Treating Allergic Rhinitis in Children: A Real- World Study Based on an Illustrious Senior Traditional Chinese Medicine Practitioner. Comb Chem High Throughput Screen 2023;26:207-23. [Crossref] [PubMed]
14. Fu QW, Liu P, Ruan Y, et al. International Evidence-Based Guidelines for Traditional Chinese Medicine Management of Allergic Rhinitis. Allergy 2025; Epub ahead of print. [Crossref]
15. Liu L. Research progress of traditional Chinese medicine in the treatment of allergic rhinitis in children. Chinese Medical Guidelines 2025;23:57-9.
16. Yang T, Luo Y, Yao L, et al. The current situation of traditional Chinese medicine diagnosis and treatment of allergic rhinitis in children. Journal of Yunnan University of Traditional Chinese Medicine 2024;47:109-12.
17. Wang H, Sun L. An exploratory study on pediatric rhinitis. Chinese Journal of Integrated Traditional and Western Medicine in Pediatrics 2023;15:425-8.
18. Yan Y, Liu R, Zhang C, et al. Clinical observation of pediatric massage in the treatment of 30 cases of pulmonary spleen deficiency in children with allergic rhinitis. Journal of Pediatrics of Traditional Chinese Medicine 2024;20:66-9.
19. Dai Q, Li T, Qian X, et al. Strategies for the prevention and treatment of allergic rhinitis in children with traditional Chinese medicine. Journal of Nanjing University of Traditional Chinese Medicine 2024;40:223-8.
20. Liang A, Ge J, Liu Z, et al. Reliability of noninvasive hyperspectral tongue diagnosis for menstrual diseases using machine learning method. Sci Rep 2025;15:6138. [Crossref] [PubMed]
21. Duan M, Zhang Y, Liu Y, et al. Machine learning aided non-invasive diagnosis of coronary heart disease based on tongue features fusion. Technol Health Care 2024;32:441-57. [Crossref] [PubMed]
22. Deng J, Dai S, Liu S, et al. Application of tongue image characteristics and oral-gut microbiota in predicting pre-diabetes and type 2 diabetes with machine learning. Front Cell Infect Microbiol 2024;14:1477638. [Crossref] [PubMed]
23. Che Q, Leng Y, Yang W, et al. Tongue Image-Based Diagnosis of Acute Respiratory Tract Infection Using Machine Learning: Algorithm Development and Validation. JMIR Med Inform 2025;13:e74102. [Crossref] [PubMed]
24. Tu S, Yin Y, Ma L, et al. Diagnosis of non-puerperal mastitis based on "whole tongue" features: non-invasive biomarker mining and diagnostic model construction. Front Cell Infect Microbiol 2025;15:1602883. [Crossref] [PubMed]
25. Zhou X, Li C, Su H, et al. Intelligent quality control of traditional chinese medical tongue diagnosis images based on deep learning. Technol Health Care 2024;32:207-16. [Crossref] [PubMed]
26. Tiryaki B, Torenek-Agirman K, Miloglu O, et al. Artificial intelligence in tongue diagnosis: classification of tongue lesions and normal tongue images using deep convolutional neural network. BMC Med Imaging 2024;24:59. [Crossref] [PubMed]
27. Wziątek-Kuczmik D, Świątkowski A, Cholewka A, et al. Thermal Imaging of the Tongue Surface as a Predictive Method in the Diagnosis of Type 2 Diabetes Mellitus. Sensors (Basel) 2024;24:2447. [Crossref] [PubMed]
28. Yang L, Dong Q, Lin D, et al. TongueNet: a multi-modal fusion and multi-label classification model for traditional Chinese Medicine tongue diagnosis. Front Physiol 2025;16:1527751. [Crossref] [PubMed]
29. Jia L, Zhang J, Zhuo R, et al. Modernizing Tongue Diagnosis: AI Integration With Traditional Chinese Medicine for Precise Health Evaluation. IEEE Access 2024;12:161670-8.
30. Zhao M, Zhou H, Wang J, et al. A new method for identification of traditional Chinese medicine constitution based on tongue features with machine learning. Technol Health Care 2024;32:3393-408. [Crossref] [PubMed]
31. Fu C, Fu Z, Yan S, et al. In-Context Multitask Learning for Few-shot Fine-tuning of Large Language Models in Traditional Chinese Medicine Tongue Diagnosis. CASSP 2025-2025 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). IEEE; 2025:1-5.
32. Jiang W, Qi J, Li X, et al. Post-infectious cough of different syndromes treated by traditional Chinese medicines: A review. Chin Herb Med 2022;14:494-510. [Crossref] [PubMed]
33. Xu J, Jiang T, Liu S. Research status and prospect of tongue image diagnosis analysis based on machine learning. Digital Chinese Medicine 2024;7:3-12.
34. Yue H, Wang X, Liu Y, et al. Enhancing Weak Feature Detection in Convolutional Analysis of Traditional Chinese Medicine Tongue Diagnosis. 2024 27th International Conference on Computer Supported Cooperative Work in Design (CSCWD). IEEE; 2024:427-32.
35. Lim J, Li J, Zhou M, et al. Machine Learning Research Trends in Traditional Chinese Medicine: A Bibliometric Review. Int J Gen Med 2024;17:5397-414. [Crossref] [PubMed]
36. Xu Q, Zeng Y, Tang W, et al. Multi-Task Joint Learning Model for Segmenting and Classifying Tongue Images Using a Deep Neural Network. IEEE J Biomed Health Inform 2020;24:2481-9. [Crossref] [PubMed]
37. Ding Q. Classification Techniques of Tongue Manifestation Based on Deep Learning. 2024 IEEE 3rd International Conference on Electrical Engineering, Big Data and Algorithms (EEBDA). IEEE; 2024: 802-810.
38. Chen Z, Zhang D, Liu C, et al. Traditional Chinese medicine diagnostic prediction model for holistic syndrome differentiation based on deep learning. Integr Med Res 2024;13:101019. [Crossref] [PubMed]
39. Subspecialty Group of Rhinology, Editorial Board of Chinese Journal of Otorhinolaryngology Head and Neck Surgery. Subspecialty Groups of Rhinology and Pediatrics, Society of Otorhinolaryngology Head and Neck Surgery, Chinese Medical Association. Zhonghua Er Bi Yan Hou Tou Jing Wai Ke Za Zhi 2022;57:392-404. [Crossref] [PubMed]
40. Wang S-c, Zhao X, Wang Y-p, et al. Guideline for the Diagnosis and Treatment of Allergic Rhinitis in Children in Chinese Medicine (Revision). Journal of Nanjing University of Traditional Chinese Medicine 2023;39:285-92.
41. Xiong Y, Yan L, Li X, et al. Hot and Cold Theory: A Personalized Medicine Approach. Adv Exp Med Biol 2021;1343:163-9. [Crossref] [PubMed]
42. Xu J, Yang Y. Traditional Chinese medicine in the Chinese health care system. Health Policy 2009;90:133-9. [Crossref] [PubMed]
43. Jin X, Guo J, Yang X, et al. Application of the Shaoyang theory in the treatment of pediatric allergic rhinitis. Clinical Research of Traditional Chinese Medicine 2023;15:63-7.
44. Bai D, Wu W, Gao Y. Traditional Chinese medicine diagnosis and treatment approaches for pediatric allergic rhinitis. New Chinese Medicine 2019;51:313-6.
45. Zhuang S, Hu L, Kong F, et al. Clinical observation of sequential therapy of traditional Chinese medicine combined with acupuncture in 42 cases of pediatric allergic rhinitis with lung qi deficiency and cold syndrome. Journal of Traditional Chinese Pediatrics 2024;20:98-102.
46. Essentials of traditional, Chinese Pediatrics. Beijing: Foreign Language Press; 1990.