Integrating machine learning into acupuncture research: a scoping review.

Objectives: Machine learning offers new tools to address the variability and subjectivity in acupuncture research. This scoping study aims to map existing literature on the use of machine learning in acupuncture research. It identifies the disease conditions most frequently targeted for machine learning-based efficacy prediction, examines the machine learning methods employed, and assesses the data inputs, methodological limitations, and existing knowledge gaps in the field. Methods: We conducted a comprehensive literature search in PubMed database using the keywords "acupuncture" and "machine learning" for publications from 2011 to 2024. Results: A total of 36 relevant articles were identified, with a notable increase after 2019. Most publications originated from China. Seventeen studies focused on predicting acupuncture efficacy, primarily for pain-related conditions. The remaining studies addressed diverse topics, including acupuncture manipulation technique detection, prescription recommendation, exploration of efficacy-related factors, acupoint sensitization prediction and specificity identification, acupuncture usage frequency prediction, investigation of acupoint-meridian conduction effects, and acupuncture robotic point localization. In efficacy prediction studies, support vector machines were the most frequently employed algorithm. Seven of the 11 studies combined Magnetic Resonance Imaging as a feature for their models, and treatment responder classification was often used as labels. Conclusion: Most studies reported encouraging predictive performance, indicating that machine learning methods can be effectively applied to acupuncture efficacy prediction. Support vector machines, in particular, demonstrated significant potential. These findings suggest that machine learning could improve the precision and efficiency of acupuncture treatments and help create more personalized and effective treatment plans. However, small sample sizes, methodological heterogeneity, inconsistent data types, and lack of standardized datasets limit model generalizability and comparability. Important gaps remain, including mechanistic understanding, long-term outcome prediction, and evaluation of clinical impact. Future research should focus on larger, multi-center studies with standardized protocols, rigorous external validation, and assessment of clinical utility to advance the integration of machine learning into acupuncture practice.