Automatic Estimation of Cervical Spine Alignment from X-ray Images Using Deep Learning Segmentation

Abstract

Reliable assessment of sagittal cervical alignment on lateral radiographs is important for diagnosing deformities and evaluating outcomes of surgical treatment, including the fusion, yet manual measurement of angles is timeconsuming and observer-dependent.We propose an automated pipeline that segments vertebral bodies C2-C7 using a UNet model and computes clinically relevant alignment metrics from endplate-derived landmarks. The approach uses segmentation as an interpretable intermediate representation and produces angle estimates directly from the predicted anatomy. Our results indicate that segmentation-based analysis can provide a practical basis for faster and more objective radiographic assessment of cervical alignment.