Skin Conductance-Based Acupoint and Non-Acupoint Recognition Using Machine Learning

Area of Science:

• Biomedical Engineering
• Traditional Chinese Medicine
• Machine Learning

Background:

• Acupoint (AP) detection currently relies heavily on manual positioning, lacking mature intelligent techniques.
• Automated AP identification is crucial for advancing clinical applications and research in Traditional Chinese Medicine.

Purpose of the Study:

• To develop and evaluate a machine learning model for recognizing acupoints (APs) and non-acupoints based on skin conductance.
• To improve the accuracy and efficiency of AP detection and localization in clinical settings.

Main Methods:

• Collected skin conductance data from Five-Shu Points and non-acupoints using wearable sensors, creating a dataset of over 36,000 samples across 12 AP types.
• Extracted electrical features from time, frequency, and nonlinear domains.
• Applied and compared machine learning algorithms including SVM, RF, KNN, NB, and XGBoost for AP/non-AP recognition.

Main Results:

• XGBoost achieved the highest recognition precision of 66.38%.
• A pairwise feature generation method was proposed to mitigate variations among AP types and individuals.
• The pairwise feature approach improved recognition precision by 7.17%.

Conclusions:

• The study successfully demonstrates systematic, automatic recognition of acupoints and non-acupoints using machine learning and skin conductance.
• This research contributes to the intelligent development of acupoint detection and Traditional Chinese Medicine theories.
• The findings support the integration of intelligent techniques into clinical practice for more precise acupoint identification.