Applications of artificial intelligence in traditional Chinese medicine tongue diagnosis: a bibliometric review of research trends and hotspots, 2015-2026.

Abstract

Objectives: This study aimed to characterize the development landscape of artificial intelligence (AI)-assisted tongue diagnosis in Traditional Chinese Medicine (TCM) by comparing publication trends, collaboration patterns, research hotspots, and emerging topics between CNKI-indexed and WoSCC-indexed literature.

Methods: Literature published between January 2015 and August 2026 was retrieved from the Web of Science Core Collection (WoSCC) and China National Knowledge Infrastructure (CNKI). Publication trends were summarized descriptively, and CiteSpace and VOSviewer were used for collaboration-network, keyword co-occurrence, clustering, timeline, burst-detection, and journal citation analyses. CNKI-indexed and WoSCC-indexed records were analysed separately because of differences in database coverage, indexing characteristics, metadata structures, and publication practices.

Results: A total of 482 studies were included, comprising 295 WoSCC-indexed and 187 CNKI-indexed records. Publication output showed an overall upward trend in both datasets over the study period. The CNKI keyword clustering network showed high clustering quality (modularity Q = 0.7929; weighted mean silhouette S = 0.9413), as did the WoSCC network (Q = 0.8000; S = 0.9239). In CNKI, the strongest keyword bursts included "deep learning" (strength = 2.95), "image segmentation" (2.69), and "transfer learning" (2.61); in WoSCC, prominent bursts included "artificial intelligence" (2.13), "image colour analysis" (2.10), and "machine learning" (1.86). CNKI-indexed literature showed prominent themes involving tongue-image objectification, segmentation, and TCM-oriented diagnostic concepts, whereas WoSCC-indexed literature showed more differentiated themes involving AI architectures, tongue-feature analysis, computational modelling, and disease-related applications. Highly cited publications in both datasets further reflected the importance of tongue-image extraction and segmentation and the subsequent expansion toward disease-oriented modelling. Across both datasets, the temporal patterns suggested a shift in research attention from conventional image processing and feature extraction toward machine learning, deep learning, and more integrated AI-based approaches.

Conclusions: The two datasets showed complementary but database-specific thematic structures in AI-assisted tongue diagnosis. Although the literature indicates a shift toward more advanced and disease-oriented AI applications, these bibliometric patterns do not establish the scientific or medical validity, diagnostic accuracy, or clinical readiness of AI-assisted tongue analysis. Further standardization, multicentre external validation, and clinically interpretable and reproducible research are needed before broader clinical application.