Degenerative cervical myelopathy (DCM) is increasing in prevalence due to population aging. Definitive diagnosis via MRI imaging may be delayed due to an equivocal clinical picture as well as limited access to affordable imaging. In this study, we proposed a deep learning approach to recognize features of degeneration at each cervical segment to identify significant spinal cord compression upon lateral radiographs. A retrospective study cohort was assembled from three study sites comprising of orthopedic clinic attendees with cervical X-rays and corresponding MRIs. The ground truth of Kang grading at each cervical segment was assessed upon mid-sagittal MRIs. A Hybrid Residual Vision Transformer was developed and validated upon each cervical segment from C2/3 to C6/7. The proposed model achieved an accuracy of 82.0%, sensitivity of 83.3%, specificity of 82.7% and ROC of 0.85 for binary classification between significant stenosis (Kang grade 2-3) and no significant stenosis (Kang grade 0-1) via independent testing (n=495). Upon whole cervical lateral radiographs, an accuracy of 90.3%, a sensitivity of 94.0% and a specificity of 80.6% was achieved upon an independent dataset (n = 113). This proof-of-concept study demonstrated that lateral radiographs may be a screening tool to predict for significant radiological cord compression in DCM.