A supervised approach towards segmentation of clinical MRI for automatic lumbar diagnosis

Abstract

Lower back pain (LBP) is widely prevalent in people all over the world. It is associated with chronic pain and change in posture which negatively affects our quality of life. Automatic segmentation of intervertebral discs and the dural sac along with labeling of the discs from clinical lumbar MRI is the first step towards computer-aided diagnosis of lower back ailments like desiccation, herniation and stenosis. In this paper we propose a supervised approach to simultaneously segment the vertebra, intervertebral discs and the dural sac of clinical sagittal MRI using the neighborhood information of each pixel. Experiments on 53 cases out of which 40 were used for training and the rest for testing, show encouraging Dice Similarity Indices of 0.8483 and 0.8160 for the dural sac and intervertebral discs respectively.