Heterogeneous Consistency Loss for Cobb Angle Estimation.

Area of Science:

• Medical Imaging
• Radiology
• Computer Vision

Background:

• Cobb angle measurement is crucial for quantifying scoliosis, a common spinal deformity.
• Current automated methods using semantic segmentation or landmark detection struggle with ambiguous vertebral appearances in X-rays.
• Robust Cobb angle estimation is essential for accurate scoliosis diagnosis and treatment planning.

Purpose of the Study:

• To develop a more robust method for automatic Cobb angle estimation in scoliosis.
• To address the limitations of existing methods in handling ambiguous vertebral features in X-ray images.
• To improve the accuracy and reliability of quantitative scoliosis assessment.

Main Methods:

• A multi-task deep learning model was proposed, simultaneously predicting semantic masks and keypoints of vertebrae.
• A heterogeneous consistency loss function was introduced to enforce consistency between predicted masks and keypoints during training.
• The model was trained and evaluated on anterior-posterior (AP) X-ray images from the AASCE MICCAI 2019 Challenge.

Main Results:

• The proposed multi-task model significantly reduced Cobb angle estimation errors compared to existing methods.
• The method achieved state-of-the-art performance on the AASCE MICCAI 2019 Challenge dataset.
• Experimental results demonstrated improved robustness in estimating Cobb angles, particularly in challenging cases with ambiguous vertebrae.

Conclusions:

• The multi-task model shows significant potential for accurate Cobb angle measurement in difficult clinical scenarios.
• Integrating this model into auxiliary clinical diagnosis systems can enhance diagnostic support for scoliosis.
• The approach offers a promising tool for more effective scoliosis assessment and subsequent treatment decisions.