Deep Learning-Based Intelligent Localization of Acupoints

Accurate acupuncture point location is crucial in Traditional Chinese Medicine (TCM) treatment. Due to individual differences in body surface characteristics, accurately locating acupuncture points can be challenging. This study proposes a deep learning-based method for the intelligent localization of human acupuncture points. We primarily use the Weizhong point (BL40) as an example to demonstrate the effectiveness of this method. BL40 is located in the middle of the knee-bone strip, but the prominence of this strip varies among individuals. To enable the intelligent positioning of BL40, we constructed and annotated a specialized dataset with the help of professional acupuncturists. We employed RTMDet and RTMPose algorithms, combined with a stepwise deep learning strategy for model training and inference. To verify the model's effectiveness, we used a test set captured by a depth camera and evaluated the model using the average Euclidean distance method. The results show that the model's offset error is 2.59 mm, which meets the error threshold for initial acupuncture point localization.