Deep learning in acupuncture: A systematic review

Objectives The systematic review aimed to comprehensively summarize evidence from the existing literature on using deep learning (DL) techniques in the practice of acupuncture. Methods Two reviewers conducted a thorough search of electronic databases, screened articles, and extracted data independently. Medline, Scopus, Embase, Web of Science, CNKI, Wanfang, CQVIP, and SinoMed were searched from 2010 to 1 August 2025. Information of the included studies, including year of publication, tasks, models, data sources, dataset size, process, main outcomes, main findings, and limitations was extracted and synthesized qualitatively. Results A total of 27 studies were included in this systematic review. The application tasks of DL models include acupoint location detection (n = 15), acupuncture manipulation analysis and classification (n = 4), disease management classification and prediction (n = 5), as well as acupuncture treatment monitoring (n = 3). All studies used self-built datasets based on public databases or self-collected datasets. The studies utilized a wide range of performance metrics, including offset error threshold, normalized mean error, mean Average Precision, Frames Per Second, Percentage of Correct Key Points, and Intersection over Union. Small data size and model inaccuracy were the two main limitations mentioned in the included studies. Conclusion The findings showcased the potential of DL models (CNN, RNN, LSTM, BERT, FNN, and YOLO variants) in detecting acupoint locations, analyzing and classifying acupuncture techniques, classifying and predicting disease management, as well as monitoring acupuncture treatment. While early applications focus primarily on standardization, newer systems demonstrate potential to enhance clinical efficacy and safety outcomes. Efforts should be directed toward addressing the challenges pertaining to data availability and model interpretability. Additionally, there is a need to establish quality appraisal tools for evaluating artificial intelligence studies.