Download 2014.10
Transcript
3.12 DWI Denoising 29 • Output Data • Non-local means filter • Discrete gaussian filter • References 3.12.1 Input Data Mandatory Input: • Diffusion weigthed image Optional Input: • Binary mask to define a denoising area. 3.12.2 Output Data • Denoised DWI: if a mask is set for denoising, all voxel excluding the masked area are set to 0. 3.12.3 Non-local means filter Denoise the input DWI using a non local means algorithm. For more details refer to [1] and [2]. Parameters: • Searchradius: defines the size of the neighborhood V (Fig. 1 b)) in which the voxels will be weighted to reconstruct the center voxel. The resulting neighborhood size is defined as 2x searchradius + 1 (e.g. a searchradius of 1 generates a neighborhood cube with size 3x3x3). • Comparisonradius: defines the size of the compared neighborhoods N (Fig. 1 b)) around each voxel. The resulting neighborhood size is defined as 2x comaprisonradius + 1 (e.g. a comparisonradius of 1 generates a neighborhood cube with size 3x3x3). • Noise variance: the variance of the noise need to be set for filtering. An estimation of the noise varinace will be implemented soon. • Rician adaption: if checked the non-local means uses an adaption for Rician distributed noise. • Joint information: if checked the whole DWI is seen as a vector image, weighting each voxels complete vector, instead of weighting each channel seperate. (This might be a bit faster, but is less accurate) Generated on Mon Dec 1 2014 12:12:38 for MITK by Doxygen