Multi-parameter scoliosis evaluation from dual-view x-rays via a local sine-based projection model.
Authors
Affiliations (5)
Affiliations (5)
- First Affiliated Hospital of Heilongjiang University of Traditional Chinese Medicine, Harbin, China, Harbin, Heilongjiang, 150040, China.
- Foshan University School of Physics and Optoelectronic Engineering, Foshan,Guangdong, Foshan, Guangdong, 528000, China.
- Foshan University School of Physics and Optoelectronic Engineering, Foshan,Guangdong, Foshan City, Guangdong, 528000, China.
- Foshan University, Foshan,Guangdong, Foshan City, Guangdong, 528000, China.
- Foshan University School of Physics and Optoelectronic Engineering, Foshan, Guangdong, Foshan, Foshan, Guangdong, 528000, China.
Abstract
To develop a robust and accurate multi-parameter assessment framework for adolescent idiopathic scoliosis (AIS) based on spinal X-ray images, overcoming the limitations of traditional Cobb angle-based evaluation. We proposed a dual-view labeling protocol and employed a Residual U-Net architecture to segment vertebral bodies in 700 paired anteroposterior and lateral X-rays. A novel Local Sine-based Spinal Projection Model (LS-SPM) was introduced to calculate vertebral inclination angles and derive consecutive Cobb angles. Additional spinal parameters, including coronal balance (T1PL-CSVL), force line (FL), and force line balance (FLB), were computed to enable comprehensive spinal evaluation. The proposed method achieved a vertebral segmentation Dice coefficient of 94.7%. The mean absolute error of the Cobb angle was 2.71°, outperforming the minimum bounding rectangle method (3.46°) and novice manual measurements (5.40°), with an intraclass correlation coefficient of 0.958. In diagnostic grading, 120 out of 142 patients were correctly classified (84.5% success rate). Longitudinal follow-up demonstrated accurate tracking of structural improvements and coronal balance recovery following conservative treatment. This approach enables automatic, precise, and comprehensive scoliosis evaluation by integrating fine-grained segmentation with geometric modeling of spinal curvature. The proposed method improves spinal assessment accuracy and facilitates longitudinal scoliosis monitoring, offering substantial clinical utility and impact in computer-aided orthopedic diagnosis.