Semi-automated segmentation of dual echo MR images

Quantitative analysis of MR images requires robust methods of segmentation. Further, it is important to be able to use standard clinical acquisition sequences to maximize the possible impact of these measures. We introduce a method of segmentation for use on conventional spin-echo MR acquisitions with two echoes. Linear combinations of proton density and T/sub 2/-weighted images enhance tissue types. These are then segmented using k-means clustering, an unsupervised classification algorithm. The segmentation occurs at two separate levels. The output of each level is combined to give a user-selected tissue type, i.e., grey matter, white matter, cerebrospinal fluid (CSF), partial volume white/grey, and partial volume CSF/grey. The segmentation is reliable and has been tested on controls as well as patients with systemic lupus erythematosus.