Deep learning algorithm enables automated Cobb angle measurements with high accuracy

Area of Science:

• Radiology
• Artificial Intelligence in Medicine
• Spinal Imaging Analysis

Background:

• Scoliosis management relies on accurate Cobb angle measurement from spinal radiographs.
• Manual measurement can be time-consuming and subject to inter-observer variability.
• Deep learning (DL) offers potential for automated and objective radiographic analysis.

Purpose of the Study:

• To evaluate the accuracy of a deep learning algorithm for automated Cobb angle measurements on full spine radiographs.
• To compare DL performance against expert manual annotations in a diverse patient cohort.

Main Methods:

• Full spine radiographs from patients over 2 years old were analyzed.
• Cobb angles were manually annotated by three expert musculoskeletal radiologists/orthopedic surgeons.
• Ground truth was established by consensus or agreement among annotators.
• A deep learning software (BoneMetrics, Gleamer) was used for automated measurements.
• Accuracy was assessed using Mean Absolute Error (MAE) compared to manual annotations.

Main Results:

• The study included 345 patients (179 pediatric, 166 adult).
• The DL algorithm achieved a Mean Absolute Error (MAE) of 2.6° for the main curvature.
• Pediatric patients showed higher accuracy with an MAE of 1.9°, compared to 3.3° in adults.

Conclusions:

• The deep learning algorithm demonstrates high accuracy in predicting Cobb angles for scoliotic patients.
• Automated DL measurements can reliably assess spinal deformities, potentially improving efficiency and consistency in clinical practice.