Classification of MR slices

An MR slice is initially segmented using FCM, then a knowledge based system utilizes a model-based recognition technique to locate successive focus-of-attention tissue areas. Specifically, qualitative models (defined by the domain knowledge) are matched with their instances in the input images. If a significant deformation is detected in a tissue, the slice is classified as abnormal. Otherwise, the system locates the next focus-of-attention tissue according to known expected tissues. This process is repeated until either a classification decision is reached or all slice tissues are labeled [8]:

1. Separate skull classes from classes of interest: white matter, gray matter, cerebro-spinal fluid (CSF) and possible tumor.
2. Locate white matter. The shape of white matter is usually deformed when large tumors exist.
3. If white matter is not deformed, use it to locate CSF. A symmetric measure of CSF determines whether there is a tumor of medium size.
4. Locate gray matter, if both white matter and CSF are believed to be normal. If there are deformations less significant in white matter, a final analysis is done at this stage with the knowledge of white matter and CSF.

Fifty-three MR slices have been tested which were acquired by a GE Advantage Tesla and a Siemens Tesla Magnetom respectively. Thirty slices are abnormal and twenty-three normal. After processing these slices, thirty-one slices were classified as abnormal and twenty-two normal. One abnormal slice with a small tumor was classified as normal, while two normal slices were classified as abnormal. One slice suffered from significant data non-uniformity while the other slice did not lie in the so-called brain "center." Slices that were classified as normal were correctly labeled.