In spinal cord injury the amount of total surviving white matter is known to be strongly related to post injury neurological functions (1). Accurate segmentation of these regions is shown to be critical in terms of developing effective treatment (1). Diffusion Tensor Imaging (DTI) has been shown to be effective in obtaining spinal cord images (2). However challenges still exist in clear separation of gray/white/cerebrospinal fluid (CSF) structures within the cord using DTI. The purpose of this study is to (1) test a semi-automatic tissue segmentation algorithm based on grow cut algorithm (GCA), to classify CSF, gray and white matter in conventional T2 weighted MRI and Diffusion Tensor Imaging (DTI) images in pediatric spinal cord injury (SCI) subjects, and (2) to compare the results of semi-automatic GCA segmentation with manually segmented spinal cord data performed on various DTI images by a board certified pediatric neuroradiologist. Results show that semi-automatic segmentation of the spinal cord using GCA was successfully implemented. Qualitatively, good separation of cord/CSF was seen in B0, CFA and FA maps (of a representative patient with SCI and a control using this GCA method. They demonstrate more homogeneous signal within the cervical spinal cord as well as greater conspicuity of the cord and surrounding CSF interface. Quantitative analysis of images segmented using GCA and manual segmentation between and within the groups showed no significant differences in CFA (p=0.1347) and FA (p=0.1442) images but B0 (p=0.0001) images showed statistically significant differences. Overall, in both the controls and subjects with SCI, quantitative and qualitative analysis showed a superior semi-automated segmentation on CFA and FA images over a B0 image the using modified GCA. Key words: Grow Cut Algorithm (GCA), Magnetic Resonance Imaging (MRI), segmentation, Diffusion Tensor Imaging (DTI), cervical spinal cord, cerebral-spinal fluid (CSF). / Bioengineering