Glenoid segmentation from CT scans based on a two-stage deep learning model for glenoid bone loss evaluation.

Researchers

Journal

Modalities

Models

Abstract

The best fitting circle drawn by CT reconstruction of the En-face view of the glenoid bone to measure the bone defect is widely used in clinical application. However, there are still some limitations in practical application, which could not achieve accurate measurement. This study aimed to accurately and automatically segment the glenoid from CT scans based on a two-stage deep learning model and quantitatively measure the glenoid bone defect.Patients who were referred to the institution between June 2018 and February 2022 were retrospectively reviewed. The dislocation group consisted of 237 patients with a history of at least 2 unilateral shoulder dislocations within 2 years. The control group consisted of 248 individuals with no history of shoulder dislocation, shoulder developmental deformity or other disease that may lead to abnormal morphology of the glenoid. All subjects underwent CT examination with a 1-mm slice thickness and a 1-mm increment, including complete imaging of the bilateral glenoids. A residual neural network (ResNet) location model and a UNet bone segmentation model were constructed to develop an automated segmentation model for the glenoid from CT scans. The dataset was randomly divided into training (201/248) and test (47/248) datasets of control group data and training (190/237) and test (47/237) datasets of dislocation group data. The accuracy of the Stage-1 (glenoid location) model, the mean intersection over union (mIoU) of the Stage-2 (glenoid segmentation) model and the glenoid volume error were used to assess the performance of the model. R-square (R2) value and Lin’s concordance correlation coefficient (CCC) were used to assess the correlation between the prediction and the gold standards.A total of 73805 images were obtained after the labeling process, and each image was composed of CT images of the glenoid and its corresponding mask. The average overall accuracy of Stage 1 was 99.28%; the average mIoU of Stage 2 was 0.96. The average glenoid volume error between the predicted and true values was 9.33%. The R2 value of the predicted and true values of the glenoid volume and glenoid bone loss (GBL) were 0.87 and 0.91. The Lin’s CCC value of the predicted and true values of the glenoid volume and GBL were 0.93 and 0.95, respectively.The two-stage model in this study showed good performance in glenoid bone segmentation from CT scans and could quantitatively measure glenoid bone loss, providing a data reference for subsequent clinical treatment.Copyright © 2023. Published by Elsevier Inc.