Neurological Symmetry Analysis Applications in Flair MRI

Abstract

<p>Neurological diseases are one of the leading causes of disability in the global population. Fluid Attenuated Inversion Recovery (FLAIR) MRI is an important imaging modality for diagnosing neurological diseases such as Alzheimer's disease (AD) and cerebrovascular disease (CVD). Analyzing characteristics of neurological diseases using FLAIR MRI provide useful neuroimaging biomarkers for diagnosis, disease monitoring/progression, and treatment planning. There is growing evidence that these neurological diseases may affect cerebral symmetry. Thus, this thesis presents the development of automated segmentation tools to facilitate clinical symmetry analysis of neuroimaging biomarkers in AD and CVD cohorts. To do this, a fully unsupervised midline estimation technique was proposed, validated against existing works, and yielded the lowest average error across all metrics: Hausdorff distance (HD) = 0.32 ± 0.23, mean absolute difference (MAD) = 1.10 ± 0.38 mm and volume difference = 7.52 ± 5.40 and 5.35 ± 3.97 ml, for left and right hemispheres, respectively. Next, white matter lesion segmentation algorithms were explored and evaluated to later develop biomarkers for CVD subjects. Of all methods explored, the SC U-Net architecture was found to be the best algorithm for WML segmentation in terms of highest Dice similarity coefficient (DSC) over most dimensions (mean DSC = 0.71 over all volumes). These automated segmentation tools enabled downstream clinical symmetry analysis. Using the midline estimation tool, hemispheric asymmetry texture biomarkers of the normal-appearing brain matter (NABM) were analyzed in a dementia cohort, and textural asymmetry was found to increase with the progression of AD. Symmetry analysis was also performed in a CVD cohort, using both the midline and WML algorithms. Significant associations between WML asymmetry and carotid disease biomarkers were found. This work demonstrates that automated algorithms developed for FLAIR MRI can be used to measure asymmetrical properties of neurological diseases, that may aid in the detection of disease, treatment planning and clinical outcomes.</p>