Acupuncture points have a positive effect on the auxiliary prevention and treatment of diseases, so medical devices such as acupuncture robots often need to combine acupuncture points to improve the treatment effect when working, however, intelligent acupoint selection technology is not yet mature, the automatic rapid and accurate positioning of acupoints is still challenging. Therefore, this paper proposes a method of back acupoint location and an evaluation index of acupoint location. First, we propose an improved Keypoint RCNN network for the preliminary location of back acupoints and introduce a channel and spatial attention mechanism module (CBAM) to improve the performance of the model. Then, we set up a posterior median line positioning method to improve the accuracy of acupoint positioning. Finally, expand and locate other acupoints according to the prior information of acupoints. According to the experimental results, the accuracy of acupoint positioning was 87.32%. After the correction of acupoint positioning, the accuracy was increased by 2.8%, which was 90.12%. In this paper, the application of depth learning in automatic location of back acupoints is realized for the first time. Only one image can be used to locate the back acupoints, with an accuracy of 90.12%.

中文翻译：

---

基于先验信息和深度学习的背部穴位定位方法


穴位对疾病的辅助预防和治疗有积极的作用，因此针灸机器人等医疗设备在工作时往往需要结合穴位来提高治疗效果，但智能选穴技术尚未成熟，自动快速选穴技术尚不成熟。而穴位的准确定位仍然具有挑战性。因此，本文提出一种背部取穴方法及取穴评价指标。首先，我们提出了一种改进的Keypoint RCNN网络用于背部穴位的初步定位，并引入通道和空间注意机制模块（CBAM）来提高模型的性能。然后，我们建立了后正中线定位方法来提高穴位定位的准确性。最后根据腧穴先验信息，拓展定位其他腧穴。实验结果显示，穴位定位的准确率为87.32%。经过穴位定位修正后，准确率提高了2.8%，达到90.12%。本文首次实现了深度学习在背部穴位自动定位中的应用。只需一张图像即可定位背部穴位，准确率达90.12%。