UCSF

The brainstem is a site of early pathology in several neurodegenerative diseases. The overall goal of this project was (a) To develop a method to segment internal brainstem structures from MP2RAGE derived images. (b) To compare the segmentations at 3 and 7 T. (c) To investigate age effects on intensities and segmentations. MP2RAGE derived T1 weighted images (UNI) and T1 relaxation maps (T1map) were obtained from two public data sets (LEMON: 50 3 T data sets, ATAG: 46 7 T data sets). The UNI and T1map images were rescaled using a linear scaling procedure and a ratio (RATIO) image calculated. The brainstem was extracted and k-mean clustering used to identify six intensity clusters from the UNI, T1map and RATIO at 3 and 7 T. Nonlinear diffeomorphic mapping was used to warp the six intensity clusters in subject space into a common space to generate probabilistic group averages/priors that were used to inform the final probabilistic segmentations at the single subject level for each field strength. The six clusters corresponded to six brainstem tissue types (three gray matter clusters and two white matter clusters and one csf/tissue boundary cluster). The quantitative comparison of the 3 and 7 T probabilistic averages showed subtle differences that affected the localization of age-associated brainstem volume losses. The segmentation approach presented here identified the same brainstem gray and white matter structures at both field strengths. Further studies are necessary to investigate how resolution and field strength contribute to the subtle differences observed at the two field strengths.