Related Experiment Video
Updated: Jul 9, 2025
09:55
Monitoring Acupuncture Effects on Human Brain by fMRI
Published on: April 8, 2010
15.5K
Hand acupuncture point localization method based on a dual-attention mechanism and cascade network model
Hao Wang1, Li Liu1,2, Ying Wang1,2
1Faculty of Mechanical Engineering and Mechanics, Ningbo University, Ningbo 315211, China.
Biomedical Optics Express
|November 29, 2023
Summary
This study introduces an improved deep learning method for precise hand acupoint localization, enhancing accuracy and reliability in traditional acupuncture. The novel approach significantly reduces localization errors, even in challenging real-world conditions.
Area of Science:
- Acupuncture and Traditional Chinese Medicine
- Medical Imaging and Computer Vision
- Deep Learning and Artificial Intelligence
Background:
- Traditional acupoint localization heavily relies on clinical experience, leading to variability.
- Existing deep learning methods show promise but require enhanced accuracy and repeatability.
- Accurate acupoint identification is crucial for effective acupuncture treatment.
Purpose of the Study:
- To develop a robust and accurate hand acupoint localization method using deep learning.
- To improve the precision and consistency of acupoint identification compared to existing techniques.
- To address challenges in complex visual environments, such as varying lighting and occlusions.
Main Methods:
- A novel method integrating dual-attention mechanisms (SE and CA) within the YOLOv5 model for hand localization.
- Utilizing K-means++ for prior box size optimization to refine hand positioning.
- Cascading a heatmap regression algorithm with HRNet as the backbone for key point detection.
- Incorporating the 'MF-cun' technique for final acupoint localization.
Main Results:
- Achieved a Frames Per Second (FPS) value of 35, ensuring real-time performance.
- Demonstrated a significantly low average offset error of 0.0269, well below the acceptable threshold.
- Reduced average offset error by over 40% compared to baseline methods.
- Validated effectiveness in complex scenarios including unequal lighting, occlusions, and diverse skin tones.
Conclusions:
- The proposed dual-attention and cascade network model offers a substantial improvement in hand acupoint localization accuracy and repeatability.
- This method provides a reliable, real-time solution for acupoint identification, overcoming limitations of traditional experience-based approaches.
- The technique is robust and adaptable to various challenging real-world conditions encountered in clinical practice.
