Semi-automated spine and intervertebral disk detection and segmentation from whole spine MR images

Abstract

Intervertebral disks are found between the vertebrae in the spine from the second cervical vertebra to the sacrum. Intervertebral disks permit various movements of the vertebral column to keep it flexible and to absorb vertical shock. The spine can be exposed to several types of compression forces that can adversely affect disk function and cause diseases or disorders, such as herniated disks. Mainly to diagnose these disorders we use Magnetic Resonance Imaging (MRI), since it's the preferred method for extracting the spinal information needed. In this paper, a Computer Aided Diagnosis (CAD) system for intervertebral disk detection and segmentation, using sagittal MR images of the whole spine has been proposed. The system consists of three major steps as follows: spine region extraction, intervertebral disk detection, and intervertebral disk segmentation. A 6th order polynomial Iterative Randomized Hough Transform (IRHT) approach is used to segment the spine sub-image from a whole spine T1/T2-weighted MR image in a relatively short time. Finally, an elliptical Hough Transform (HT) is used to detect the intervertebral disks from the segmented spine image. The proposed system used a whole spine images dataset consisting of 31 spine T1/T2 MR images. The accuracy measures of intervertebral disk detection and disk segmentation are 94.65% and 90%, respectively. The results show that the proposed system is robust and accurate for spine segmentation and intervertebral disks detection with high performance. Finally, the proposed system is a useful tool for radiologists in detecting as well as segmenting intervertebral disk in whole spine MR images.

Keywords

• T2-weighted MR image
• Intervertebral disk
• Polynomial IRHT
• Whole spine
• Detection
• Segmentation