Related Experiment Video
Updated: Apr 10, 2026
06:45
Automated Joint Space Detection Improves Bone Segmentation Accuracy
Published on: November 28, 2025
283
Multicenter deep learning for multi-abnormality screening on hip radiographs: development, external validation, and
Shenghao Xu1, Chaohui Guo2, Qibo Xu3
1Department of Orthopedics, The Second Hospital of Jilin University, Changchun, Jilin, China; Joint International Research Laboratory of Ageing Active Strategy and Bionic Health in Northeast Asia of the Ministry of Education, Jilin University, Changchun, Jilin, China.
Journal of Advanced Research
|April 8, 2026
Summary
A new deep learning (DL) model can screen multiple hip abnormalities from pelvic radiographs, improving diagnostic accuracy and aiding orthopedic surgeons. This AI tool assists in identifying high-risk cases for faster review.
Area of Science:
- Artificial Intelligence in Radiology
- Deep Learning for Medical Imaging
- Orthopedic Diagnostics
Background:
- Hip abnormalities cause significant pain and functional impairment, with missed diagnoses being common due to clinical workload and interobserver variability.
- Current deep learning (DL) models often focus on single pathologies, limiting their use for comprehensive hip condition screening.
Purpose of the Study:
- To develop and validate a DL model for the simultaneous screening of multiple hip conditions.
- Utilized a large, multicenter dataset of pelvic radiographs for model development and testing.
Main Methods:
- A ResNet-50 based DL model was trained and validated on 25,908 hips, incorporating enhancements for improved performance.
- The model classified eight hip conditions: normal, osteoarthritis, osteonecrosis, femoral neck fracture (FNF), intertrochanteric fracture (ITF), developmental dysplasia, total hip arthroplasty, and internal fixation.
- External validation was performed on 4,600 hips, and the model's performance was compared against orthopedic surgeons.
Main Results:
- The DL model achieved high accuracy (93.93% internal, 90.11% external) and macro-F1 scores (90.66% internal, 87.29% external).
- Demonstrated high sensitivity for acute fractures (FNF: 95.61%, ITF: 95.31%).
- The model outperformed residents and attendings, and model assistance improved surgeon accuracy and inter-rater agreement.
Conclusions:
- An externally validated DL model for automated screening of multiple common hip abnormalities has been developed.
- The system is suitable for triage workflows, flagging high-risk cases for expedited review.
- The DL model can serve as a supportive second reader in high-volume or resource-limited settings.