AIR-Net: Acupoint image registration network for automatic acupoint recognition and localization

Acupoint localization is integral to various Traditional Chinese Medicine (TCM) practices, including acupuncture, moxibustion, and massage. We introduce a cutting-edge atlas-based image registration framework that leverages deep neural networks for the automatic identification and localization of acupoints. This innovative approach incorporates both local and global features within a Transformer network, meticulously designed to assimilate comprehensive human body morphology and detailed acupoint data. Further bolstering our methodology is an expansive dataset consisting of 89,951 pairs of images, each meticulously annotated with acupoint labels to facilitate precise localization. By integrating body contours with specific acupoint indicators, our Transformer-based network sets a new precedent in acupoint recognition precision. Preliminary experiments demonstrate the efficacy of our proposed framework, achieving an impressive accuracy of over 90%— a significant improvement over current state-of-the-art solutions. This notable enhancement in acupoint localization underscores our method's potential to substantially elevate the precision and reliability of TCM clinical practices.