Accurate Cobb Angle Estimation via SVD-Based Curve Detection and Vertebral Wedging Quantification.

Area of Science:

• Orthopedics
• Medical Imaging
• Artificial Intelligence

Background:

• Adolescent idiopathic scoliosis (AIS) is a prevalent spinal deformity.
• Current Cobb angle measurements for AIS severity have significant observer variability.
• Existing automated methods lack the complexity to handle diverse AIS presentations.

Purpose of the Study:

• To develop a novel deep learning framework for accurate AIS assessment.
• To preserve the anatomical reality of vertebral wedging in progressive AIS.
• To introduce a new metric for predicting AIS progression.

Main Methods:

• A deep learning framework combining HRNet backbone and Swin-Transformer modules.
• Singular Value Decomposition (SVD) for flexible detection of diverse scoliosis patterns.
• Utilized 630 full-spine anteroposterior radiographs with dual-rater annotation.

Main Results:

• Achieved 83.45% diagnostic accuracy and 2.55° mean absolute error.
• Demonstrated exceptional generalization on out-of-distribution cases.
• Introduced the Vertebral Wedging Index (VWI) showing prognostic correlation with curve progression.

Conclusions:

• The deep learning framework offers a more accurate and reliable method for AIS assessment.
• The novel VWI metric shows significant potential for predicting AIS curve progression.
• This approach supports early detection, personalized treatment, and monitoring of AIS.