Deformable Image Registration in the Analysis of Multiple Sclerosis

Medical Imaging;Image Registration;Image Segmentation;Distortion Correction;Population Analysis;Magnetic Resonance Imaging;Optical Coherence Tomography;Multiple Sclerosis;Electrical Engineering

Chen, MinPrince, Jerry L. ;

Johns Hopkins University

关键词: Medical Imaging;    Image Registration;    Image Segmentation;    Distortion Correction;    Population Analysis;    Magnetic Resonance Imaging;    Optical Coherence Tomography;    Multiple Sclerosis;    Electrical Engineering;   

Others  :  https://jscholarship.library.jhu.edu/bitstream/handle/1774.2/37894/CHEN-DISSERTATION-2015.pdf?sequence=1&isAllowed=y

瑞士|英语

来源: JOHNS HOPKINS DSpace Repository

【 摘 要 】

In medical image analysis, image registration is the task of finding corresponding features in two or more images, and using them to solve for the transformation that best aligns the images. Knowing the alignment allows information, such as landmarks and functional metrics, to be easily transferred between images, and allows them to be analyzed together.This dissertation focuses on the development of deformable image registration techniques for the analysis of multiple sclerosis (MS), a neurodegenerative disease that damages the myelin sheath of nervous tissue. MS is known to affect the entire central nervous system (CNS), and can result in the loss of sensorimotor control, cognition, and vision. Hence, the four primary contributions of this dissertation are on the development and application of deformable image registration in the three areas of the CNS that are most currently studied for MS -- the spinal cord, the retina, and the brain.First, for spinal cord magnetic resonance imaging (MRI), an approach is presented that uses deformable registration to provide atlas priors for automatic topology-preserving segmentation of the spinal cord and cerebrospinal fluid. The method shows high accuracy and robustness when compared to manual raters, and allows spinal cord atrophy to be analyzed on large datasets without manual segmentations. Second, for spinal cord diffusion tensor imaging, a pipeline is presented that uses deformable registration to correct for susceptibility distortions in the images. The pipeline allows for accurate computation of spinal cord diffusion metrics, which are shown to be significantly correlated with clinical measures of sensorimotor function and disability levels. Third, for optical coherence tomography (OCT) of the retina, a deformable registration technique is presented that constrains the transformation to follow the OCT acquisition geometry. 3D voxel-based analysis using the algorithm found significant differences between healthy and MS cohorts in regions of the retina that is consistent with previous findings using 2D analysis. Lastly, for brain MRI, a multi-channel registration framework is presented that can use distance transforms and image synthesis to improve registration accuracy. Together, these techniques have enabled several types of analysis that were previously unavailable for the study of MS.