Abstract
Segmentation of brain MR images plays an important role in longitudinal investigation of developmental, aging, disease progression changes in the cerebral cortex. However, most existing brain segmentation methods consider multiple time-point images individually and thus cannot achieve longitudinal consistency. For example, cortical thickness measured from the segmented image will contain unnecessary temporal variations, which will affect the time related change pattern and eventually reduce the statistical power of analysis. In this paper, we propose a 4D segmentation framework for the adult brain MR images with the constraint of cortical thickness variations. Specifically, we utilize local intensity information to address the intensity inhomogeneity, spatial cortical thickness constraint to maintain the cortical thickness being within a reasonable range, and temporal cortical thickness variation constraint in neighboring time-points to suppress the artificial variations. The proposed method has been tested on BLSA dataset and ADNI dataset with promising results. Both qualitative and quantitative experimental results demonstrate the advantage of the proposed method, in comparison to other state-of-the-art 4D segmentation methods.
Highlights
With the rapid development of MR imaging technology and its widespread use, large number of MR images are obtained for clinical studies
The preprocessing of the input longitudinal images includes the following steps: (1) intensity correction of each image using N3 [26]; (2) to avoid bias, the input serial images are rigidly aligned onto an atlas space and the group-mean image can be constructed by averaging all rigidly aligned images; (3) skull stripping [27] and removing the cerebellum using in-house tools on the group-mean image; (4) warping the brain mask of the group-mean image back to the each time-point image space based on the inverted transform matrix and removing the non-brain using the warped brain mask
Results on Simulated Data To generate simulated images with longitudinal deformations, we used Atrophy Simulation Package, which can simulate the atrophy by matching the Jacobian of the simulated deformation to the desired volumetric changes, subject to smoothness and topology preserving constraints employed in the algorithm [30]
Summary
With the rapid development of MR imaging technology and its widespread use, large number of MR images are obtained for clinical studies. The resulting cortical thickness measured from the segmented image will contain unnecessary temporal variations, which will affect the time related change pattern and eventually reduce the statistical power of analysis To this end, several 4D segmentation methods were proposed in recent years to address this problem by including the temporal constraint between time-points in the segmentation process [6,7,8,9,10]. In [6], the authors proposed a temporally consistent and spatially adaptive longitudinal MR brain image segmentation algorithm based on FANTASM, referred to as CLASSIC, which aims at obtaining accurate measurements of rates of change of regional and global brain volumes from serial MR images. The cortical surfaces at each time point are deformed to achieve longitudinal cortical surface reconstruction
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
