Validation of Nonlinear Spatial Filtering to Improve Tissue Segmentation of MR Brain Images

Intensity-based tissue segmentation of MR brain images is facilitated if the image noise can be effectively reduced without removing significant image detail. Signal to noise ratio can be increased by averaging multiple acquisitions of the same subject after proper geometric alignment, but this penalizes acquisition time. In this paper we evaluate the effect of nonlinear spatial filtering of single scans prior to tissue classification. Spatial filtering is performed iteratively by nonlinear intensity diffusion to suppress noise in homogeneous tissue regions without smoothing across tissue boundaries. We validate the impact on segmentation accuracy using simulated MR data with known ground-truth, demonstrating that the performance obtained with spatial filtering of single scans is comparable to that of averaging multiple coregistered scans. The performance can be further improved by tuning the filter parameters towards optimal segmentation accuracy.