Abstract
Measurement of brain structures could lead to important diagnostic information and could indicate the success or failure of a certain pharmaceutical drug. We have developed a totally unsupervised technique that segments and quantifies brain structures from T2 dual echo MR images. The technique classified four different tissue clusters in a scatter plot (air, CSF, brain, and face). Several novel image-processing techniques were implemented to reduce the spread of these clusters and subsequently generate tissue based T2 windows. These T2 windows encompassed all the information needed to segment and subsequently quantify the corresponding tissues in an automatic fashion. We have applied the technique on nineteen MR data sets (16 normal and 3 Alzheimer diseased [AD] patients). The measurements from the T2 window technique differentiated AD patients from normal subjects. The mean value of the %CSF from total the brain was %29.2 higher for AD patients from the %CSF for normal subjects. Furthermore, the technique ran under 30 seconds per data set on a PC with 550 MHz dual processors.