Download or read book Development and Validation of Automatic Tools for Segmentation of White Matter Hyperintensities written by Mahsa Dadar and published by . This book was released on 2018 with total page pages. Available in PDF, EPUB and Kindle. Book excerpt: "Automatic methods for segmentation of various tissues and pathologies are critical for systematic studies of the brain to investigate changes that occur in different components of the phenomenon under study. White matter hyperintensities (WMHs) are one of the major components of small-vessel disease in aging and Alzheimer's disease (AD) populations that need to be assessed and monitored to estimate the vascular disease burden. In this thesis, a new fully automatic technique is proposed for segmenting WMHs from multiple contrasts of magnetic resonance (MR) brain images. The proposed segmentation technique uses a machine learning classification scheme by combining a set of intensity and location features obtained from multi-contrast MR sequences, namely T1w, T2w, proton density (PD) and fluid attenuated inversion recovery (FLAIR) images and a linear or nonlinear classifier to detect WMHs. The segmentations are performed in the native space of the optimal contrast (e.g. FLAIR or T2w) to avoid the blurring caused by resampling the images, especially since these images generally have relatively thick slices (3-5 mm). The classifiers are then trained on the training dataset with manually segmented labels. The performance of the classifiers is assessed using Dice Kappa values as the primary outcome measure and through a 10-fold cross validation scheme.Using the developed tool, the WMHs were segmented using different combinations of input image contrasts (i.e. T1w+T2w+PD, T1w+FLAIR, T1w) to assess the performance of the classifiers and the contribution of each of the contrasts in detecting WMHs. The question of interest was whether the WMHs loads obtained from segmentations based only on T1w images can be used as accurate estimates of the actual WMH loads. To assess this, the volumetric correlation of WMH loads in different brain lobes as well as correlation with age and cognitive measures were compared to investigate the effectiveness of each contrast in providing WMH load estimates that are highly correlated with aging and cognitive scores. The assessments revealed that the best Dice Kappa values are obtained while using the optimal FLAIR and T2w/PD contrasts. Classifications based solely on T1w images tend to undersegment the WMHs, only detecting the brightest of these lesions on FLAIR and T2w/PD images. However, the WMH loads obtained from T1w segmentations were still able to provide high correlations with age and cognitive scores. Finally, using the developed tool, baseline WMHs were segmented in an early stage Parkinson's disease (PD) database as well as age matched healthy controls. Using longitudinal clinical assessments and cortical thickness measures, we studied the relationship between baseline WMHs and future cognitive decline and cortical thinning. PD subjects with high WMH loads were found to present with more future cognitive decline and cortical thinning in comparison with (i) PD subjects with low WMH loads and (ii) age matched control subjects with high WMH loads. These findings show that the existence of WMHs affects PD patients differently from controls. " --