Fully automated axial plane segmentation of cervical spinal cord using U-Net in MR scans

Abstract

The spinal cord belongs to the central nervous system and is responsible for maintaining vital functions. It transmits all relevant sensory and motor signals from the body to the brain and from the brain to the body. Therefore, clinical examination and monitoring of the spinal cord and diagnosis of possible disease symptoms in this region require very sensitive assessments. Thanks to advances in imaging technology and improved image quality, diseases and disorders of the spinal cord can be more easily visualized and monitored. In this study, the cross-sectional area (CSA) and cerebrospinal fluid (CSF) regions of the cervical spinal cord are automatically segmented using T2- weighted MR images scanned from the axial plane using the UNet deep learning architecture. A new cervical spinal cord dataset was created using data from Akdeniz University Hospital. In the experimental studies, a Dice similarity coefficient (DSC) score of 0.9144 was achieved using the proposed U-Net architecture for fully automated segmentation of the cervical spinal cord. Therefore, it can be concluded from the DSC scores that the cervical spinal cord is segmented with high performance using the proposed U-Net method.