Document Type : Original Paper
Department of Medical Physics, Faculty of Medicine, Mashhad University of Medical Sciences, Mashhad, Iran
Department of Neurology, Ghaem Hospital, Faculty of Medicine, Mashhad University of Medical Sciences, Mashhad, Iran
Mashhad University of Medical Sciences
Introduction: Accurate segmentation of brain tissue in magnetic resonance imaging (MRI) is an important step in the analysis of brain images. There are automated methods used to segmentation the brain and minimize the disadvantages of manual segmentation, including time consuming and misinterpretations. These procedures usually involve a combination of skull removal, bias field correction, and segmentation. Therefore, segmented tissue quality assessment segmentation of gray matter (GM), white matter (WM), and cerebrospinal fluid (CSF) is required for the analysis of neuroimages.
Material and Methods: This paper presents the performance evaluation of three automatic methods brain segmentation, fluid and white matter suppression [FSL, Freesurfer (FreeSurfer is an open source package for the analysis and visualization of structural, functional, and diffusion neuroimaging data from cross-sectional and longitudinal studies) and SPM12 (Statistical Parametric Mapping)]. Segmentation with SPM12 was performed on three tissue probability maps: i) threshold 0.5, ii) threshold 0.7 and iii) threshold 0.9. In order to compare and evaluate the automatic methods, the reference standard method, i.e., manual segmentation, was performed by three radiologists.
Results: Comparison of GM, WM and CSF segmentation in MR images was performed using similarities between manual and automatic segmentation. The similarity between the segmented tissues was calculated using diagnostic criteria.
Conclusion: Several studies have examined the classification of GM, WM, and CSF using software packages. In these studies, different results have been obtained depending on the type of method and images used and the type of segmented tissues. In this study, the evaluation of the segmentation of these packages with reference standard method is performed. The results can help users in selecting an appropriate segmentation tool for neuroimages analysis.