Related Experiment Video
Updated: Aug 27, 2026
Swin-PSAxialNet: An Efficient Multi-Organ Segmentation Technique
Published on: July 5, 2024
Three-dimensional automatic segmentation and morphometric analysis of scapula computed tomography images using
Qiong Fang1, Meimei Liu1, Jun Xu1
1School of Basic Medicine, Anhui Institute of Medicine, Hefei, China.
Objectives:
To develop and validate a method based on Swin-U-Net transformers for three-dimensional (3D) automatic segmentation of scapula computed tomography (CT) images and automated measurement of scapular morphometric parameters.
Methods:
CT scans from 106 healthy adults were retrospectively collected. The Swin-U-Net transformers model performed 3D scapular segmentation and automatic localization of 13 anatomical landmarks. A reference coordinate system was established from these landmarks to measure glenoid version, inclination, apex angle, height, width, and lateral acromial extension (LEA). LEA variations were analyzed according to age, sex, laterality, and acromion morphology (Bigliani classification).
Results:
The model achieved a Dice coefficient of 0.95 ± 0.02 and a Hausdorff distance of 2.81 ± 0.63 mm; landmark localization error averaged 1.90 ± 0.50 mm. Measured glenoid parameters closely matched manual measurements. LEA averaged 29.67 ± 0.24 mm, with larger values on the right side and in type III (hooked) acromion morphologies (p<0.05). Age and sex had no significant effect. LEA showed strong bilateral correlation (r=0.904, p<0.001).
Conclusions:
The Swin-U-Net transformers-based method enables accurate, automated 3D scapular segmentation and quantitative morphometric analysis, providing a reliable tool for anatomical research, clinical assessment, and surgical planning of the shoulder.
