Segmentation of spinal cord in the pediatric spinal Diffusion Tensor MR Imaging

Classification and segmentation of small structures such as spinal cord is extremely challenging. In this paper, a multi stage segmentation algorithm is proposed and tested to accurately and reliably segment the spinal canal and spinal cord from the background in the pediatric spinal Diffusion Tensor MR images. First, median filter and image compression methods were applied to mitigate the amplitude of the noise and improve the homogeneity of the image. Next, mathematical morphological processing was applied to segment and label the regions attributed to the spinal canal. These segmented regions were classified into the spinal canal and background using a Euclidean metric obtained by centroid coordinates of segmented regions in the volumetric DTI data. Finally, Otsu thresholding technique was applied to extract cord region from spinal canal. Segmentation accuracy, sensitivity, specificity and spatial overlap index were examined as performance measurements. The quantitative measurements represent the effectiveness of the proposed method.