Automating Cobb Angle Measurement for Adolescent Idiopathic Scoliosis using Instance Segmentation

Area of Science:

• Medical Imaging
• Orthopedics
• Artificial Intelligence

Background:

• Scoliosis, a spinal deformity affecting millions, is typically diagnosed in childhood.
• Current Cobb angle measurement for scoliosis is manual, time-consuming, and prone to observer variability.
• Automating Cobb angle measurement can improve diagnostic accuracy and efficiency.

Purpose of the Study:

• To develop and validate a machine learning-based method for automated Cobb angle measurement in scoliosis.
• To improve the reliability and reduce the time required for scoliosis assessment.
• To overcome the inter- and intra-observer variance inherent in manual measurements.

Main Methods:

• Utilized an instance segmentation model (YOLACT) to segment vertebrae in X-ray images.
• Employed minimum bounding boxes to track decisive vertebral landmarks.
• Calculated Cobb angles based on the extracted landmark coordinates.

Main Results:

• Achieved a Symmetric Mean Absolute Percentage Error (SMAPE) of 10.76% for Cobb angle measurement.
• Demonstrated high accuracy, with over 94% of estimated Cobb angles having an error less than ten degrees.
• Showcased reliable performance in both vertebra localization and Cobb angle calculation.

Conclusions:

• The proposed machine learning method offers a reliable and accurate alternative to manual Cobb angle measurement for scoliosis.
• Automating this process with AI can significantly enhance the efficiency and consistency of scoliosis diagnosis.
• This approach holds promise for improving patient care and outcomes in pediatric orthopedics.