Fully Automated Measurement of Cobb Angles in Coronal Plane Spine Radiographs

Area of Science:

• Orthopedics
• Radiology
• Medical Imaging
• Artificial Intelligence in Medicine

Background:

• Scoliosis is a complex spinal deformity involving lateral and rotational curvature.
• Current gold-standard Cobb angle measurement in scoliosis assessment has significant interrater variability (up to 10°).
• Need for objective and reproducible methods in spinal deformity assessment.

Purpose of the Study:

• To evaluate a fully automated artificial intelligence (AI) method for measuring Cobb angles in scoliosis.
• To assess the performance and agreement of the AI model against a radiologist-established reference standard.
• To investigate the AI's accuracy in identifying end vertebrae for Cobb angle calculation.

Main Methods:

• Utilized 196 AP/PA full-spine radiographs for analysis.
• Established a reference standard using median Cobb angle measurements from four radiologists.
• Employed an AI-based software (IB Lab SQUIRREL, version 1.0) for automated Cobb angle measurements.

Main Results:

• AI demonstrated 88.58% accuracy in selecting end vertebrae compared to radiologists.
• AI measurements showed minimal bias (mean difference 0.16° ± 0.35°) compared to the reference standard.
• High interrater reliability with an Intraclass Correlation Coefficient (ICC) of 0.97 between AI and reference standard.

Conclusions:

• The AI model achieved excellent accuracy in automated Cobb angle measurements.
• The AI demonstrated strong performance in end vertebrae determination.
• This AI tool shows potential as a reliable aid in clinical scoliosis assessment and research.