Unsupervised Lumbar IVD Localization and Segmentation using GFMM and Boundary Refined Region Growing Techniques

Abstract

Low Back Pain is caused because of Lumbar Intervertebral Disc (IVD) degeneration, and it is one of the most suffered problems by a large population. in this paper, the lumbar IVD is automatically localized and segmented using Gabor Filter with Mathematical Morphology and novel Boundary Refined Region Growing techniques, respectively. an MRI dataset is used to validate the suggested approach, consisting of 180 IVDs from 30 subjects. Initially, the Gabor Filter with Mathematical Morphology and Support Vector Machine with Local Binary Pattern techniques are used in localizing the lumbar IVD. in comparison to performance, Gabor Filter with Mathematical Morphology localized with 100% accuracy. In contrast, SVM localized with 89.4% for the precision range of 2mm. the Gabor Filter with Mathematical Morphology attained an accuracy of 96.9% for the 0.6mm precision range, which is comparatively higher than the accuracy of SVM for the 2mm precision range. Then the segmentation is preceded by the novel Boundary Refined Region Growing technique on the lumbar IVD image localized by Gabor Filter with Mathematical Morphology, achieving a better Dice Similarity Index, sensitivity, and specificity of 86.2%, 92%, and 99%, respectively.