Figure 1b. Images and graphs illustrate how SEGMENT, a fully automated algorithm, uses (a) T2- and (b) intermediate-weighted images to plot all pixels as (c) a global histogram in a T2-weighted (x axis) versus intermediate-weighted (PD; y axis) feature space. SEGMENT then uses an algorithm with maximum likelihood criteria to classify pixel clusters as brain, CSF, CSF partially volumed with skull (cluster not shown in figure), or other (nonbrain). Next, for all pixels classified as brain, SEGMENT uses a three-dimensional local-contrast algorithm to separate GM and WM pixels, which creates (d) final GM-, WM-, and CSF-segmented images. For qualitative comparison, manual tracings of T2- and intermediate-weighted pairs of images (eg, see tracings in a and b) were overlaid onto SEGMENT-created images (tracing in d). (e) Validation of SEGMENT was accomplished by means of statistical comparison with results from manual tracing; analyses showed that GM and WM measures obtained by using SEGMENT and manual-tracing methods correlated to 98.2% and 96.1%, respectively.