Massage Acupoint Positioning Method of Human Body Images Based on Transfer Learning

Chao Zhang, Qian Wu, Ju Wang, Liyan Yang, Hongxia ZhangTraitement 2023 · DOI

Abstract

Traditional Chinese massage therapy is a very popular method to stay healthy, which regulates body balance, alleviates fatigue, and prevents diseases by massaging specific acupoints. Although computer vision has been increasingly applied in traditional Chinese medicine, related study of acupoint positioning is still insufficient. The existing acupoint positioning methods mainly rely on manual labeling and rule matching, which often require a large amount of manual intervention with limited accuracy. Therefore, this study proposed a massage acupoint positioning method of human body images based on transfer learning. The massage acupoint meridian and collateral positioning principle of human body images was presented. Using the integrated deep belief network model as a pre-trained model, a feasible transfer learning model was established through fine-tuning and feature mapping. The experimental results verified that the proposed method was effective.

Introduction

Traditional Chinese Medicine (TCM) massage relies heavily on the precise location of acupoints. Historically, this has been a manual process or one governed by rigid geometric rules, both of which struggle with the natural variability of human anatomy. While computer vision has made strides in medical imaging, its application to acupoint positioning remains limited by two core challenges: the scarcity of high-quality, labeled datasets and "domain drift," where models fail to generalize across different skin tones, lighting, or pathological conditions such as swelling or redness.

Methodology

The authors propose a multi-stage pipeline that moves from coarse body localization to fine-grained acupoint detection, leveraging advanced pathfinding and deep learning techniques.

Meridian Pathfinding via Fast Marching Method (FMM)

To find the actual paths of meridians (the channels through which energy flows), the system uses the Fast Marching Method (FMM) to solve the eikonal equation. This approach is superior to standard grid-based algorithms like Dijkstra because it better handles the continuous nature of image data.

Flowchart of the Fast Marching Method (FMM) algorithm for meridian pathfinding.

The algorithm iteratively calculates arrival times for points, moving from "Alive" sets to "Trial" and "Far" sets. To ensure the path stays centered on anatomical structures, the authors incorporated "inertia features" into the cost function, guiding the model along the body's natural contours.

Illustration of acupoint positioning utilizing inertia feature guidance to maintain path consistency.

Transfer Learning and Domain Adaptation

Instead of training from scratch, the study utilizes an integrated Deep Belief Network (DBN) as a pre-trained model. To solve the domain drift problem, they implement Multiple Kernel Maximum Mean Discrepancy (MK-MMD). This technique measures and minimizes the distribution differences between the training data (source domain) and real-world test data (target domain), allowing the model to adapt to different individuals and environments.

Results and Discussion

The proposed method was benchmarked against Random Forest, ResNet, and U-Net across three distinct categories: normal samples, abnormal (pathological) samples, and samples with background interference.

  • Accuracy: On difficult "abnormal" samples featuring redness or inflammation, the model maintained 82.61% accuracy, far outperforming ResNet (15.92%) and U-Net (42.58%).
  • Stability: The model showed high stability with the lowest median error (1.251) and variance (0.539) among tested architectures.
  • Efficiency: Prediction times ranged from 2.46 to 5.38 seconds, making it viable for near-real-time clinical applications.
  • Optimization: Fine-tuning the pre-trained model specifically for acupoint tasks reduced the positioning error range from 1.5–2.4 down to 1.1–1.9.

Conclusion

This research provides a robust framework for automated acupoint positioning by combining transfer learning with sophisticated pathfinding optimization. While the current results are promising, future work could involve 3D body scanning to account for the depth and curvature of the human torso and the integration of this software into physical massage robots or augmented reality (AR) systems for practitioners.

Summary updated

Abstract

Traditional Chinese massage therapy is a very popular method to stay healthy, which regulates body balance, alleviates fatigue, and prevents diseases by massaging specific acupoints.Although computer vision has been increasingly applied in traditional Chinese medicine, related study of acupoint positioning is still insufficient.The existing acupoint positioning methods mainly rely on manual labeling and rule matching, which often require a large amount of manual intervention with limited accuracy.Therefore, this study proposed a massage acupoint positioning method of human body images based on transfer learning.The massage acupoint meridian and collateral positioning principle of human body images was presented.Using the integrated deep belief network model as a pre-trained model, a feasible transfer learning model was established through fine-tuning and feature mapping.The experimental results verified that the proposed method was effective.Relevant research results provide useful references for research in related fields.
BibTeX
@inproceedings{zhang2023massage,
  author = {Chao Zhang and Qian Wu and Ju Wang and Liyan Yang and Hongxia Zhang},
  title = {{Massage Acupoint Positioning Method of Human Body Images Based on Transfer Learning}},
  booktitle = {Traitement},
  year = {2023},
  doi = {10.18280/ts.400636},
}
Figure