AcuSim: A Synthetic Dataset for Cervicocranial Acupuncture Points Localisation

Abstract

The locations of acupuncture points (acupoints) differ among human individuals due to variations in factors such as height, weight and fat proportions. However, acupoint annotation is expert-dependent, labour-intensive, and highly expensive, which limits the data size and detection accuracy. In this paper, we introduce the "AcuSim" dataset as a new synthetic dataset for the task of localising points on the human cervicocranial area from an input image using an automatic render and labelling pipeline during acupuncture treatment. It includes a creation of 63,936 RGB-D images and 504 synthetic anatomical models with 174 volumetric acupoints annotated, to capture the variability and diversity of human anatomies. The study validates a convolutional neural network (CNN) on the proposed dataset with an accuracy of 99.73% and shows that 92.86% of predictions in validation set align within a 5mm threshold of margin error when compared to expert-annotated data. This dataset addresses the limitations of prior datasets and can be applied to applications of acupoint detection and visualization, further advancing automation in Traditional Chinese Medicine (TCM).

Fig. 1

Illustration of the workflow of dataset construction. (a) shows the general process of dataset creation, whereas the data annotation and segmentation approach for the AcuSim creation is shown in (b).

Fig. 2

Examples of synthetic human models for domain-randomized training data. Diverse variations (body sizes, hairstyles, eye states) are applied as primary variables. Skin tones are identity-specific but not controlled variables. These variations close the simulation to real gaps, enhancing the synthetic data realism to ensure it accurately reflects real-world scenarios.

Fig. 3

Subset of 3D rendered models and images captured from multiple perspectives including 4 pitch angles (30° steps) and 36 rotation angles (10° steps).

Fig. 4

Determining acupoint occlusion by comparing the surface mesh’s normal vectors with the camera’s orientation. The main figure shows a side view of the cervicocranial model with acupoints DU18 and DU22, where angles α and β measure the deviation between the camera’s orientation and normal vectors. Subfigure 1 and 2 provide enlarged views of DU18 and DU22, respectively, with normal vectors and viewing angles.

Fig. 5

This image illustrates the annotated cervicocranial acupoints, displaying key facial landmarks across multiple angles and perspectives. The model is shown rotating 360° around the vertical axis, with views from the front, top-down, high angle, and low angle.

Fig. 6

Multitask deep neural network for facial feature extraction and acupoint prediction including names and coordinates.

Fig. 7

The Loss and Accuracy Comparison in Training and Validation.

Fig. 8

Violin plot for illustration of the performance of our trained model in predicting the acupoint’s locations when comparing to the expert labelled data and the margin of error.

Fig. 9

Quantitative error analysis of acupoint localization accuracy. A subset of 10 synthetic models (F_Brisen to M_Zeno) and 13 acupoints (BLL to ST2) were analyzed. Mean Euclidean distance differences are shown for: (a) deviations between dataset acupoint coordinates and model predictions; (b) deviations from expert annotations; (c) deviations between the Dataset-Prediction errors and the normal vectors of the camera and body mesh.

Fig. 10

Subfigure (a) and (b) represent the correlation analysis maps within the Dataset-Prediction Deviation and Dataset-Expert Deviation respectively.