Automatic Cobb angle measurement method based on vertebra segmentation by deep learning.

Area of Science:

• Medical Imaging
• Computer Vision
• Biomedical Engineering

Background:

• Accurate Cobb measurement is critical for scoliosis diagnosis and treatment.
• Manual Cobb angle measurement is prone to observer variability, impacting progression evaluation.

Purpose of the Study:

• To develop a fully automatic Cobb measurement method to overcome the limitations of manual assessment.
• To enhance the accuracy of vertebra segmentation for improved Cobb angle estimation.

Main Methods:

• An improved U-shaped network incorporating Inception and Res Blocks for multi-scale feature extraction.
• Integration of Convolutional Block Attention Module (CBAM) into the U-Net bottleneck to refine feature importance.
• Development of an efficient automatic Cobb angle measurement algorithm based on segmented vertebrae.

Main Results:

• The proposed U-Net model achieved superior vertebra segmentation performance, with Dice coefficients significantly outperforming state-of-the-art methods.
• The automated Cobb angle measurement yielded a mean absolute error (CMAE) of 2.45°, substantially lower than the 5-7° error of manual measurements.
• Experimental validation on 75 spinal X-ray images confirmed the accuracy and reliability of the proposed method.

Conclusions:

• The enhanced U-shaped network effectively improves vertebra segmentation accuracy for scoliosis assessment.
• The proposed automatic Cobb measurement method offers a reliable and precise alternative to manual techniques, reducing observer variability in clinical settings.