All slices entering this system have been previously classified as abnormal and are known to contain gadolinium-enhanced glioblastoma multiforme-tumor. And since this is an extension of our previous work, slices used in this system have been previously ``pre-processed'' by those systems in order to generate certain knowledge required by this system. Details on how those facts are generated can be found in [1,2,3].
The first piece of knowledge inherited from the pre-processing stages is the separation of clusters containing extra-cranial pixels from clusters containing intra-cranial pixels. This was the first step performed by the knowledge-systems in [1,2,3] and creates two sets of facts. These fact sets can separate clusters into Group 1 and Group 2 clusters respectively. The groups are mutually exclusive.
Identification of white-matter clusters is the second piece of knowledge generated in pre-processing. After all extra-cranial tissues are removed, white matter occupies the lowest end of the T2 spectrum.
Finally, other facts concerning the brain slice in question, including the centroid of the intra-cranial region (the brain center) and its lengths of the major and minor axes will becomes important facts.
In creating the first ``recluster mask'' (which controls which pixels are passed to FCM) the knowledge gained during pre-processing plays its most important role in aiding focus-of-attention. Since the extra-cranial tissues and white matter have already been identified and are not of interest in tumor segmentation, they can be used to generate an image mask with reclustering performed on the remaining pixels. An example of the mask can be seen in Figure 3.
Figure 3: Building the Initial Recluster Mask. Note pathology originally clustered
into extra-cranial tissues. Image (a) is the original segmented image;
(b) is the intra-cranial tissues; the dark region in
(c) is pixels remaining after white matter removal.