Fractional Segmentation of White Matter [chapter]

Simon K. Warfield, Carl-Fredrik Westin, Charles R. G. Guttmann, Marilyn Albert, Ferenc A. Jolesz, Ron Kikinis
1999 Lecture Notes in Computer Science  
Abnormalities in the white matter of the brain are common to subjects with multiple sclerosis and Alzheimer's disease. They also develop in normal, asymptomatic, subjects and appear more frequently with age. Clinically, it is interesting to be able to differentiate between different disease states and to find markers which allow early diagnosis. Conventional spin echo (CSE) magnetic resonance imaging (MRI) is sensitive to these white matter changes and has frequently been applied to their
more » ... Previous approaches to investigate white matter abnormalities have often been reported to have difficulty distinguishing between normal gray matter and abnormal white matter due to their similar appearance in MRI. Earlier methods have also often generated binary classifications, reporting white matter as either normal or abnormal. We have developed a new approach which first identifies the region of white matter using a template moderated spatially varying classification, and then estimates the degree of white matter abnormality present at each voxel of the white matter. This fractional segmentation allows us to preserve the heterogeneous characteristics of white matter abnormalities and to investigate both focal and diffuse white matter damage. We compute, from the fractional segmentation, a white matter spectrum showing the different levels of white matter damage present in each subject. We applied this automated image segmentation method to over 996 MRI scans of subjects affected by multiple sclerosis, 72 normal aging subjects and 29 scans of subjects with Alzheimer's disease. We investigated the ability to characterize these different subject groups based upon tissue volumes determined by spatially varying classification, and by the fractional segmentation of the white matter of each patient.
doi:10.1007/10704282_7 fatcat:xqvv2wnoinhkjgfoilylrtkknm