Conquering the Cobb Angle: A Deep Learning Algorithm for Automated, Hardware-Invariant Measurement of Cobb Angle on...

Area of Science:

• Radiology and Imaging
• Orthopedics
• Artificial Intelligence in Medicine

Background:

• Scoliosis affects over 8% of US adults, diagnosed via manual Cobb angle measurement on radiographs.
• Manual Cobb angle measurements are time-consuming, hindering surgical planning and patient monitoring.
• Accurate and efficient scoliosis assessment is crucial for patient management.

Purpose of the Study:

• To develop and validate an automated pipeline for measuring Cobb angles in scoliosis patients.
• To assess the accuracy and robustness of a deep learning algorithm for Cobb angle measurement.
• To evaluate the algorithm's performance across diverse patient demographics and imaging types.

Main Methods:

• A retrospective study utilizing the SpineTK architecture trained on 1310 anterior-posterior images.
• The pipeline was validated and tested on internal (n=460) and external (n=161) datasets.
• Deep learning algorithms, specifically convolutional neural networks, were employed for automated Cobb angle measurement.

Main Results:

• The algorithm achieved high accuracy, with less than 2° error (ICC=0.96) compared to expert radiologists.
• Measurements were rapid, completed in under 0.5 seconds.
• Performance remained consistent across various factors including surgical hardware, age, sex, BMI, scoliosis severity, and image type.

Conclusions:

• The developed automated algorithm provides rapid, accurate, and robust Cobb angle measurements for scoliosis.
• This deep learning tool shows significant potential for improving scoliosis progression monitoring and clinical workflow.
• The algorithm's consistency across diverse patient characteristics supports its clinical applicability.