Evaluation of Local Thresholding Algorithms for Segmentation of White Matter Hyperintensities in Magnetic Resonance Images of the Brain

White matter hyperintensities are distinguished in magnetic resonance images as areas of abnormal signal intensity. In clinical research, determining the region and position of these hyperintensities in brain MRIs is critical; it is believed this will find applications in clinical practice and will support the diagnosis, prognosis, and therapy monitoring of neurodegenerative diseases. The properties of hyperintensities vary greatly, thus segmenting them is a challenging task. A substantial amount of time and effort has gone into developing satisfactory automatic segmentation systems. In this work, a wide range of local thresholding algorithms has been evaluated for the segmentation of white matter hyperintensities. Nine local thresholding approaches implemented in ImageJ software are considered: Bernsen, Contrast, Mean, Median, MidGrey, Niblack, Otsu, Phansalkar, Sauvola. Additionally, the use of other local algorithms (Local Normalization and Statistical Dominance Algorithm) with global thresholding was evaluated. The segmentation accuracy results for all algorithms, and the parameter spaces of the best algorithms are presented.