자괴공명 뇌 영상 분석 방법이 종적 뇌 영상 분석에 미치는 영향에 관한 연구

Abstract

As increasing importance of longitudinal image analysis in studies of aging, brain development and neurodegenerative diseases, the importance of accurate measurement of brain magnetic resonance image on the study of neurodegenerative disorder is being emphasized. Image processing in MRI-based neuro-anatomical studies is often performed in a cross-sectional manner where each time point is evaluated independently. For this reason, cross-sectional method generally tends to incur higher measurement error than expected. For these reason, several methods have been proposed in the past decades for longitudinal image analysis. We propose to compare the effects of each step in longitudinal analysis and evaluate how much variability can be reduced by analyzing structural volume and cortical thickness changes in longitudinal data. We separate longitudinal image processing methods into two portion that consists of registration and segmentation for longitudinal analysis. First, the registration step includes the creation of subject template. Second portion is longitudinal segmentation step using information transform from template to subject. To study the effects of these two steps, we used three framework including cross-sectional and longitudinal pipeline (template registration and longitudinal segmentation) and one longitudinal method that used template registration and cross-sectional segmentation process. We can evaluate the effects of registration when compare the cross-sectional methods and longitudinal method that used template registration and cross-sectional segmentation process. Furthermore, we can estimate the segmentation effects that improved for longitudinal analysis when compare the two longitudinal methods. To evaluate the how much variability can be reduced, we use scan-rescan dataset (HCP) and longitudinal dataset(ADNI). ADNI dataset contain AD group and NC group data. All experiments are performed by FreeSurfer software version 5.3. The result presents that both steps of the longitudinal framework reduce the variability in scan-rescan and longitudinal data set. Moreover, the result shows that longitudinal method that used template registration and cross-sectional segmentation process has a higher effect on structural volume change studies. And the result shows that the longitudinal pipeline that used both longitudinal registration and segmentation process has more influence on cortical thickness variation.