Abstract
MRI of cerebral white matter may show regions of signal abnormalities. These changes may be associated with hypertension, inflammation, or ischemia, as well as altered brain function. The goal of this work has been to construct computational atlases of white matter lesions that represent both their severity as well as the frequency of their occurrence in a population to achieve a better classification of white matter disease. An atlas is computed with a pipeline that uses 4T FLAIR and 4T T1-weighted (T1w) brain images of a group of subjects. The processing steps include intensity correction, lesion extraction, intra-subject FLAIR to T1w rigid registration, and seamless replacement of lesions in T1w images with synthetic white matter texture. Subsequently, the T1w images and lesion images of different subjects are registered non-rigidly to the same space. The decrease in T1w intensities is used to obtain severity information. Atlases were constructed for two groups of subjects, elderly normal controls or with mild cognitive impairment, and subjects with cerebrovascular disease. The lesion severities of the two groups have a significant statistical difference with the severity in the atlas of cerebrovascular disease being higher.
Collapse