Automatic Identification of the Scapular Border to Increase the Efficiency of Data Processing for the Freehand Three-Dimensional Ultrasound System

摘要

The ability to visualize the scapula in three dimensions (3D) is necessary for the evaluation of scapular movement. The scapula plays an important role in upper extremity function as it provides a stable base for shoulder movement and enables optimal shoulder complex function. We previously developed a custom freehand ultrasound (FUS) system for purposes of evaluating bone movement, which is a relatively unexplored application as it pertains to shoulder biomechanics. Our system was developed to create a reconstructed scapular border in 3D space, from points of interest in two-dimensional (2D) ultrasound images, and determine scapular rotations. We found high reliability in evaluating scapular kinematics in static postures with our 3D FUS system. However, we are currently limited to manual detection of the scapular border in the ultrasound images, which is very time consuming. Steps are needed to enhance the FUS system to include automatic detection and increase efficiency. For this study, we have developed a program, capable of automatically identifying and tracking the scapula in 2D ultrasound images, to be integrated into our 3D FUS system. Selected coordinates identified as the scapular border by our automated program were compared to previous manual selections to validate its accuracy and reliability. Using intraclass correlation coefficients, we found substantial to excellent inter-rater reliability (agreement between the automated and manual point selections). The semi-automated point selection program reduces the data processing time required for identification of the spine and medial border of the scapula in our ultrasound images by over 50%. Our results suggest that this proposed program is a viable method for automatically identifying and tracking the scapular border in 2D ultrasound images. Further study on image pre-processing prior to future application of this automated program should be conducted to further improve the accuracy of our algorithm. In conclusion, point selection is necessary for 3D reconstruction of the scapular border and this automation ultimately enhances our FUS system by increasing the efficiency of our point selection process. Access to 3D scapular models plays several roles ranging from detection of shoulder pathologies to assessing the effectiveness of interventions or preventative measures for shoulder injuries.