Artificial intelligence in cervical spinal imaging: a review of deep learning segmentation models

Cervical spinal cord segmentation in T2-weighted MRI is essential for diagnosing conditions like degenerative myelopathy. Manual methods are time-consuming and inconsistent. To evaluate the clinical readiness of deep learning (DL) models, particularly convolutional neural networks (CNNs), for cervical spinal cord segmentation. A structured literature review of studies from 2015-2025 was conducted using Scopus. Eligible studies applied DL to cervical spinal cord segmentation in T2-weighted MRI and reported accuracy metrics and clinical applicability. DL models such as U-Net and RET-Net achieved Dice scores exceeding 0.90, showing expert-level performance. Compared to manual and semi-automated methods, DL models offered faster, more consistent, and scalable segmentation. DL-based segmentation methods demonstrate high accuracy and clinical promise for cervical spine imaging. Further validation across diverse populations is needed to support widespread clinical integration.