Deep Learning Approach for Hand Acupoint Localization Combining Reflex Zones and Topological Keypoints
Yi Zheng, Shuai Chen, Qiang He · Procedia Computer Science · 2024
This paper builds on traditional Chinese medicine theory and deep learning techniques, integrating classical acupoint localization methods with a modern computational approach. This study addresses the prevalent inaccuracies in current acupoint detection algorithms by incorporating the topological relationships between acupoints and their reflex zones into the localization process. Improvements were made to the traditional MediaPipe keypoint detection algorithm by mapping the spatial and topological relationships of 21 key points identified on the hand to corresponding acupoints. This enhanced method combines acupoint reflex zone and topology-based acupoint detection strategies. After merging the derived hand acupoint maps with reflex zone along the channel dimensions, the YOLOv8 algorithm is trained to produce precise coordinates for 16 specific acupoints. Compared to the original MediaPipe acupoint localization approach, the refined algorithm demonstrates significant improvements in accuracy across various acupoints, achieving a maximum precision increase of 3.218 millimeters. This advancement underscores the potential of integrating traditional acupoint knowledge with deep learning techniques to enhance the effectiveness and accessibility of acupuncture treatments.