Segmentation methods for MRI human spine images using thresholding approaches

Abstract

Computer-Aided Diagnosis (CAD) in MRI image processing can assist experts in detecting abnormality in human spine image efficiently. The manual process of detecting abnormality is tedious, hence the use of CAD in this field is helpful to increase the diagnosis efficiency. The segmentation method is an important and critical process in CAD that could affect the accuracy of the MRI spine image’s overall diagnosis. There are various segmentation methods commonly used in CAD. One of the methods is segmentation using thresholding. Thresholding approaches divide the area of interest by identifying the threshold values that can separate the image into desired grayscale levels based on its pixel’s intensity. This study focuses on investigating the optimal approach in segmenting lumbar vertebrae on the MRI images. The steps involved in this study include pre-processing (normalization), segmentation using local and global thresholding, neural network classification, and performance measurement. 20 images are used to evaluate and compare the segmentation methods. The effectiveness of the segmentation method is measured based on the performance measurement technique. This preliminary study shows that local thresholding outperforms the global thresholding approach with an accuracy of 91.4% and 87.7%.