TIGP (BIO)—Deep Learning in Musculoskeletal Ultrasound Imaging

Abstract

Musculoskeletal ultrasound imaging involves examination of a variety of anatomical structures relevant to musculoskeletal disorders, including muscles, tendons, ligaments, fascia, nerves, vessels, joints, and bony structures such as spinous processes. In additional to assessing the anatomical structures using static images, musculoskeletal ultrasound imaging provides real-time visualization of the structure movement as the subject performs dynamic maneuver, revealing pathology unseen in static conditions. The evaluation of disease severity usually involves manual measurement of the changes in the morphological characteristic of the abnormal structure. However, this demands tons of human labor and data reliability and reproducibility may be compromised by the variation among the operators. A promising solution to save the human’s effort and reduce measurement inconsistency is to employ deep-learning based approaches to assist the pathology classification and automate the biometric measurement. In this talk, representative applications of deep learning in musculoskeletal ultrasound images will be addressed, including our work in median nerve segmentation. Limitations and potential solutions to current challenges will also be discussed.

Download