AcuSim: A Synthetic Dataset for Cervicocranial Acupuncture Points Localisation.

Area of Science:

• Biomedical Engineering
• Medical Imaging
• Traditional Chinese Medicine

Background:

• Acupoint locations vary individually due to anatomical differences.
• Current manual acupoint annotation is costly, time-consuming, and subjective, limiting data and accuracy.
• Automated acupoint detection is crucial for advancing Traditional Chinese Medicine (TCM).

Purpose of the Study:

• To introduce the AcuSim dataset for synthetic acupoint localization in the cervicocranial region.
• To address limitations of existing datasets in terms of size, cost, and accuracy.
• To facilitate automated acupoint detection and visualization.

Main Methods:

• Developed an automatic rendering and labeling pipeline to create synthetic RGB-D images.
• Generated 63,936 images and 504 synthetic anatomical models with 174 annotated volumetric acupoints.
• Validated a convolutional neural network (CNN) on the AcuSim dataset.

Main Results:

• The CNN achieved 99.73% accuracy in acupoint localization on the AcuSim dataset.
• 92.86% of predictions aligned within a 5mm error margin compared to expert annotations.
• The synthetic dataset effectively captures anatomical variability for diverse human anatomies.

Conclusions:

• The AcuSim dataset provides a scalable and accurate resource for acupoint detection research.
• Automated methods using this dataset can significantly enhance precision and efficiency in TCM.
• This work paves the way for wider adoption of automated techniques in acupuncture treatment.