TY - JOUR
T1 - Automated multiclass bone segmentation using deep learning
T2 - implications for templating in radial head replacement
AU - Velasquez Garcia, Ausberto R.
AU - Yang, Linjun
AU - Nishikawa, Hiroki
AU - Fitzsimmons, James S.
AU - Wentworth, Adam J.
AU - Morris, Jonathan M.
AU - Taunton, Michael J.
AU - O'Driscoll, Shawn W.
N1 - Copyright © 2026 Journal of Shoulder and Elbow Surgery Board of Trustees. Published by Elsevier Inc. All rights reserved.
PY - 2026
Y1 - 2026
N2 - Background: Preoperative three-dimensional (3D) templating can improve surgical accuracy in anatomic-press-fit radial head arthroplasty (RHA). However, current imaging segmentation methods used for templating are time-consuming and prone to variability. This study aimed to train and validate an no-new U-Net (nnU-Net) deep learning model to automate multiclass bone segmentation for RHA templating. We hypothesized that the nnU-Net model would achieve high accuracy in segmenting the upper extremity bones thereby supporting 3D bone templating in RHA. Methods: A total of 93 upper extremity computed tomography (CT) scans met the eligibility criteria. Ground-truth segmentation was performed by a trained orthopedic surgeon and reviewed by a radiologist and an engineer to ensure accuracy. The nnU-Net model was trained and evaluated using the Dice similarity coefficient and Hausdorff distance to measure overlap and segmentation accuracy against manual segmentations. The 3D bone models derived from the nnU-Net model and manual segmentation were compared through mean surface distance and root mean squared error to assess the surface variation between the bone models. The average time on segmenting each CT was compared. Results: The nnU-Net achieved high segmentation accuracy with Dice similarity coefficient values of 0.99 for the humerus, 0.98 for the ulna, and 0.96 and 0.95 for the cortical and noncortical radii, respectively. The mean surface distance remained below 0.2 mm for all bone classes. The mean root mean squared error values were consistent at 0.2 mm across all bones. Segmentation time averaged 3 min per scan compared to 78 min for manual segmentation, with consistent performance across gender, arm side, and CT slice thickness. Discussion and Conclusion: This deep learning model provides a fast and reliable solution for multiclass bone segmentation and demonstrates high accuracy in segmenting cortical and noncortical regions, which are essential for RHA templating. The accuracy was consistent with clinical needs and fits below the sizing intervals of commercially available prostheses. This supports its potential utility for 3D preoperative planning in RHA, despite its inability to capture cartilage. This approach demonstrates clinical feasibility for improving efficiency and precision in templating radial head replacement surgery.
AB - Background: Preoperative three-dimensional (3D) templating can improve surgical accuracy in anatomic-press-fit radial head arthroplasty (RHA). However, current imaging segmentation methods used for templating are time-consuming and prone to variability. This study aimed to train and validate an no-new U-Net (nnU-Net) deep learning model to automate multiclass bone segmentation for RHA templating. We hypothesized that the nnU-Net model would achieve high accuracy in segmenting the upper extremity bones thereby supporting 3D bone templating in RHA. Methods: A total of 93 upper extremity computed tomography (CT) scans met the eligibility criteria. Ground-truth segmentation was performed by a trained orthopedic surgeon and reviewed by a radiologist and an engineer to ensure accuracy. The nnU-Net model was trained and evaluated using the Dice similarity coefficient and Hausdorff distance to measure overlap and segmentation accuracy against manual segmentations. The 3D bone models derived from the nnU-Net model and manual segmentation were compared through mean surface distance and root mean squared error to assess the surface variation between the bone models. The average time on segmenting each CT was compared. Results: The nnU-Net achieved high segmentation accuracy with Dice similarity coefficient values of 0.99 for the humerus, 0.98 for the ulna, and 0.96 and 0.95 for the cortical and noncortical radii, respectively. The mean surface distance remained below 0.2 mm for all bone classes. The mean root mean squared error values were consistent at 0.2 mm across all bones. Segmentation time averaged 3 min per scan compared to 78 min for manual segmentation, with consistent performance across gender, arm side, and CT slice thickness. Discussion and Conclusion: This deep learning model provides a fast and reliable solution for multiclass bone segmentation and demonstrates high accuracy in segmenting cortical and noncortical regions, which are essential for RHA templating. The accuracy was consistent with clinical needs and fits below the sizing intervals of commercially available prostheses. This supports its potential utility for 3D preoperative planning in RHA, despite its inability to capture cartilage. This approach demonstrates clinical feasibility for improving efficiency and precision in templating radial head replacement surgery.
KW - artificial intelligence
KW - Basic Science Study
KW - Computer Modeling
KW - Deep learning
KW - nnU-Net
KW - preoperative planning
KW - radial head arthroplasty templating
KW - upper extremity bone segmentation
UR - https://www.scopus.com/pages/publications/105036349047
U2 - 10.1016/j.jse.2026.02.005
DO - 10.1016/j.jse.2026.02.005
M3 - Article
C2 - 41720251
AN - SCOPUS:105036349047
SN - 1058-2746
JO - Journal of Shoulder and Elbow Surgery
JF - Journal of Shoulder and Elbow Surgery
ER -