Abstract
This paper shows a novel medical image segmentation suitable for the human brain magnetic resonance(MR)volumetric images. The method is based on fuzzy information granulation. The concept is introduced by Zadeh. He considers that the information including imprecision and uncertainty consists of some fuzzy granules. Fuzzy information granulation is defined as deriving the granules from the information. The human brain MR volumetric images consist of several parts; the gray matter, white matter, cerebrospinal fluid and so on. We treat their parts as the fuzzy granules. By developing the fuzzy matching technique to aid the fuzzy information granulation, we can segment the brain region from the MR volume. An experiment is done on 50 human brain MR volumes. The error ratio, on the average, is 2.3%, compared to the manually segmented volumes by a medical doctor.
