Mixture modeling applied to the partial volume effect in MRI data

In statistical image classification, each voxel is normally assigned to one of the classes in the training set. However, this is not generally an adequate model of reality since the signal detected in one voxel may be derived from two or more different texture types. This paper presents a statistical model for voxels composed of a mixture of multiple tissue types. These voxels are called mixels. The probability density function of mixture is simulated in the case of mixels consisting of two pure tissue types when tissue intensities are represented by gaussian distribution functions. The histogram of MRI data intensities upon from an SPGR acquisition sequence is then fitted. Finally, the authors specifically discuss the mixture modeling expectation to the classification of soft tissues in MR images