Noise reduction and brain tissue classification in MR images

Abstract
The effect of Gaussian noise and noise reduction techniques on subsequent tissue classification in MR (magnetic resonance) images was investigated. For image restoration, a nonlinear Gaussian window filter was used before classifying the multispectral MR acquisitions from the head. The Bayesian classifier was based on J. Haslett's (1985) contextual method. The classification experiments demonstrated that it was possible to recognize normal brain tissue types and some pathological lesions. Preprocessing with the Gaussian window filter did not improve the tissue classification results in the case of images with normal noise level. The robustness of the classifier against noise (which is shown to be Gaussian) was explored. Noisy input severely affected the classification results. However, the number of misclassifications was clearly reduced after applying a Gaussian window filter with a fixed smoothing parameter Author(s) Lundervold, A. Norwegian Comput. Center, Oslo, Norway Godtliebsen, F.

This publication has 3 references indexed in Scilit: