A knowledge based automatic region of interest (ROI) segment of cervical cord diffusion tensor imaging

Diffusion MR imaging technique has been developed in past decade to permit the detection of tissue water molecular diffusion at microscopic dimension and has been widely used to investigate the spinal cord parenchyma. The commonly used hand-drawn region of interest (ROI)-based segmentation in diffusion tensor imaging(DTI) analysis is highly labor-intensive and user-dependent. In this study, we aim to develop an automatic template (auto-template) based on anatomy knowledge and computational intelligence to extract the regional diffusion anisotropy pattern of cervical spinal cord. A total of 16 healthy volunteers were recruited in this study. Eleven axial diffusion tensor MR images covering C1 to C7 of cervical spinal cord were taken with single-shot spin-echo echo-planar imaging sequence on a 3T MR system. The fractional anisotropy (FA) value of anterior, lateral, posterior column of white matter and gray matter was measured using hand-drawn ROI and knowledge based computational intelligence method respectively. Knowledge based template showed the FA value in the anterior (0.737±0.009), lateral (0.827±0.011), posterior (0.854±0.007) column of white matter and gray matter (0.493±0.009) with higher inter-rater agreement than hand-drawn ROI method. The result suggested that knowledge based template is a convenient tool to extract the diffusion data in various parts of the cervical spinal cord with high accuracy and inter-rater reliability.