Deep learning model for measuring the sagittal Cobb angle on cervical spine computed tomography.

Area of Science:

• Radiology
• Medical Imaging
• Artificial Intelligence

Background:

• Cervical spine deformities require accurate angle measurement for diagnosis and treatment.
• Manual measurement of the sagittal Cobb angle can be subjective and time-consuming.

Purpose of the Study:

• To develop and validate a deep learning (DL) model for automated measurement of the cervical spine sagittal Cobb angle using computed tomography (CT).

Main Methods:

• Two DL models (VB-Net based) were developed for cervical vertebra segmentation and key-point detection.
• Automated sagittal Cobb angle calculation using four-points and line-fitting methods.
• Performance evaluation against manual measurements from two doctors using PCK, ICC, Pearson correlation, MAE, and Bland-Altman plots on internal (991 patients) and external (112 patients) datasets.

Main Results:

• The DL model achieved high accuracy in key-point detection (PCK 78-100%).
• The line-fitting method showed strong agreement with the reference standard (internal ICC=0.97, MAE=3.23°; external ICC=0.80, MAE=4.68°).
• Both automated methods demonstrated good correlation (r=0.94 internal, r=0.974 external for line-fitting).

Conclusions:

• The developed DL model accurately measures the cervical spine sagittal Cobb angle on CT.
• The line-fitting method within the DL model offers superior consistency and reduced error compared to manual measurements.