Related Experiment Video
Updated: Mar 22, 2026
A Spine Robotic-Assisted Navigation System for Pedicle Screw Placement
Published on: May 11, 2020
Scalable and Robust Artificial Intelligence for Spine Alignment Assessment: Multicenter Study Enabled by Real-Time
Guilin Chen1, Nan Meng2, Yipeng Zhuang2
1Department of Orthopaedic Surgery, State Key Laboratory of Complex Severe and Rare Diseases, Beijing Key of Big Data Innovation and Application for Skeletal Health Medical Care, Key Laboratory of Big Data for Spinal Deformities, Peking Union Medical College Hospital, Peking Union Medical College and Chinese Academy of Medical Sciences, Beijing, 100730, China.
A new data transformation method enhances AI accuracy for adolescent idiopathic scoliosis (AIS) spinal assessment across hospitals. This improves cobb angle prediction and disease grading, making AI more reliable in diverse clinical settings.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Spine Deformities
Background:
- Artificial intelligence (AI) shows potential for automating spinal alignment assessment in adolescent idiopathic scoliosis (AIS).
- AI model performance often decreases across multiple medical centers due to variations in imaging protocols and data.
- This variability can compromise clinical diagnosis and treatment decisions for AIS.
Purpose of the Study:
- To develop a real-time, plug-and-play data transformation method to improve AI model robustness against data heterogeneity in radiographs.
- To enhance the performance of deep learning models for AIS assessment across multiple medical centers.
Main Methods:
- A retrospective multicenter study included 3899 full-spine radiographs from 7 hospitals.
- A novel pixel intensity-based data transformation method standardized image contrast and brightness.
- The method was integrated into the SpineHRNet+ AI model for evaluating cobb angle (CA) prediction and severity classification accuracy and robustness.
Main Results:
- The data transformation method significantly reduced contrast variability between datasets.
- The enhanced SpineHRNet+ achieved consistent CA predictions across external datasets (mean error within 4°) with R² > 0.90.
- Sensitivity and negative predictive value for disease severity grading improved to 90.18% and 93.16%, respectively.
Conclusions:
- The data transformation approach effectively improved AI accuracy and robustness for multicenter AIS assessments.
- The method's real-time processing and preservation of anatomical integrity enhance clinical practicality.
- This enables scalable and reliable AI applications in diverse healthcare environments for AIS management.
