Automatic Calculation of Cervical Spine Parameters Using Deep Learning: Development and Validation on an External...

Area of Science:

• Radiology
• Medical Imaging
• Deep Learning

Background:

• Accurate measurement of cervical spine parameters is crucial for diagnosing and managing spinal conditions.
• Manual measurement of these parameters from radiographs can be time-consuming and prone to inter-observer variability.

Purpose of the Study:

• To develop and validate a deep learning model for automated calculation of important cervical spine parameters from lateral cervical radiographs.

Main Methods:

• Retrospective analysis of two datasets from different institutions (1498 images for training, 79 for validation).
• A deep learning model was trained to predict parameters including T1 slope, C7 slope, C2-C7 angle, C2-C6 angle, Sagittal Vertical Axis (SVA), C0-C2, Redlund-Johnell distance (RJD), cranial tilting (CT), and craniocervical angle (CCA).
• Model performance was evaluated using median absolute errors against ground truth measurements.

Main Results:

• The model achieved low median absolute errors for angle measurements (e.g., 1.66° for T1 slope, 1.56° for C7 slope) and distance measurements (e.g., 0.55 mm for SVA, 0.47 mm for RJD).
• Performance was consistent across different parameters, indicating high accuracy.

Conclusions:

• A deep learning model was successfully developed for accurate prediction of cervical spine parameters from lateral cervical radiographs.
• The model demonstrated robustness and high generalizability on an external validation set from a different institution.