Deep learning automates Cobb angle measurement compared with multi-expert observers

Area of Science:

• Medical Imaging Analysis
• Spinal Deformity Assessment
• Artificial Intelligence in Healthcare

Background:

• Scoliosis, a common spinal deformity, requires accurate measurement of the Cobb angle for diagnosis and management.
• Manual Cobb angle measurement is time-consuming, labor-intensive, and suffers from significant inter- and intraobserver variability.
• Existing automated methods often lack interpretability, posing challenges for clinical adoption.

Purpose of the Study:

• To develop and validate a fully automated software for precise Cobb angle measurement in scoliosis.
• To improve the interpretability and reproducibility of scoliosis assessments.
• To provide a reliable tool for enhanced clinical diagnosis and patient care.

Main Methods:

• Integration of a deep neural network for spine region detection and segmentation.
• Automated spine centerline identification and localization of maximally tilted vertebrae.
• Direct visualization of Cobb angles on original radiographic images.

Main Results:

• The automated algorithm achieved a mean deviation of 4.17 degrees in Cobb angle measurements, outperforming the manual average intra-reader discrepancy of 5.16 degrees.
• Intraclass correlation coefficients (ICC) exceeded 0.96, and Pearson correlation coefficients were above 0.944, indicating strong agreement with expert assessments.
• The software demonstrated superior measurement reliability and robustness compared to manual methods.

Conclusions:

• The developed algorithm offers a highly accurate and reproducible method for Cobb angle measurement in scoliosis.
• The automated system enhances interpretability and consensus with expert readers, promising significant clinical utility.
• This tool has the potential to improve the accuracy of scoliosis diagnosis and management, ultimately benefiting patient care.