Transformer Based Real Time Musculoskeletal Anatomical Structure Detection in Clinical Use

Medical ultrasound imaging is known as a non-invasive and radiation-free real-time imaging method, making it an indispensable technique for clinical use, especially in musculoskeletal medicine. However, due to the complexity of the body's internal anatomical structures, it is challenging for physicians to rapidly and accurately interpret images to identify each structure for further diagnosis and treatment planning. To address these challenges, we developed a real-time labeling system of anatomical structures for musculoskeletal ultrasound using a transformer-based model, RT-DETR. We selected the 15 most crucial anatomical structures in the human forearm and shoulder as our identification targets. We achieved an overall precision of 0.834, recall of 0.822, mAP50 of 0.855, mAP50:95 of 0.526, and an F1-score of 0.82. Moreover, the real-time annotation results also show high precision in annotating each structure's classes and boundaries, indicating that our model is well-suited for the task and has the potential to be applied in clinical settings.