N3 Inhomogeneity Correction

N3 Inhomogeneity Correction
This preprocessing algorithm corrects for shading artifacts often seen in MRI. The heart of the algorithm is an iterative approach that estimates both a multiplicative bias field and a distribution of true tissue intensities. Referred to as nonparametric intensity nonuniformity normalization (N3), this method makes no assumptions about the kind of anatomy present in a scan and is robust, accurate, and fully automatic.
Input Types

You should be able to apply this algorithm to 2D and 3D MRI images.

N3 Parameters

Signal threshold

Treats those voxels of intensity that are lower than this threshold value as background.

The default value is 1.0.

Max number iterations

Specifies the maximum number of iterations allowed before the process is terminated.

The default value is 50.

End tolerance

Specifies convergence when all membership functions (of the fuzzy C cluster segmentation) over all pixel locations j change by less than this tolerance value between two iterations. The default value is 0.001.

Field distance

Specifies the characteristic distance over which the field varies. The distance between adjacent knots in B-spline fitting with at least 4 knots going in every dimension. The automatic default is one third of the distance (resolution * extents) of the smallest dimension.

Subsampling Factor

Specifies the value by which data is subsampled to a lower resolution in estimating the slowly varying nonuniformity field. It reduces sampling in the finest sampling direction by the shrink factor. Sampling in other directions is reduced only if reduced sampling in direction of finest resolution is less than current sampling. The default value is 4.

Kernel FWHM

Specifies the width of deconvolution kernel F (a Gaussian) used to sharpen the V(pdf of log(v), where v is the measured signal) histogram. Larger values give faster convergence while smaller values give greater accuracy. The default value is 0.01.

Wiener Filter Noise

Indicates the noise used in Wiener filter.

Automatic histogram thresholding

Computes a threshold value through histogram analysis that should remove just the background from the computation. Selecting this check box disables the Signal threshold box, the Whole image, and VOI regions.

Example Usage

Example input image.

Output image, after N3 Correction.

Output inhomogeneity field.

Signal threshold	Treats those voxels of intensity that are lower than this threshold value as background. The default value is 1.0.
Max number iterations	Specifies the maximum number of iterations allowed before the process is terminated. The default value is 50.
End tolerance	Specifies convergence when all membership functions (of the fuzzy C cluster segmentation) over all pixel locations j change by less than this tolerance value between two iterations. The default value is 0.001.
Field distance	Specifies the characteristic distance over which the field varies. The distance between adjacent knots in B-spline fitting with at least 4 knots going in every dimension. The automatic default is one third of the distance (resolution * extents) of the smallest dimension.
Subsampling Factor	Specifies the value by which data is subsampled to a lower resolution in estimating the slowly varying nonuniformity field. It reduces sampling in the finest sampling direction by the shrink factor. Sampling in other directions is reduced only if reduced sampling in direction of finest resolution is less than current sampling. The default value is 4.
Kernel FWHM	Specifies the width of deconvolution kernel F (a Gaussian) used to sharpen the V(pdf of log(v), where v is the measured signal) histogram. Larger values give faster convergence while smaller values give greater accuracy. The default value is 0.01.
Wiener Filter Noise	Indicates the noise used in Wiener filter.
Automatic histogram thresholding	Computes a threshold value through histogram analysis that should remove just the background from the computation. Selecting this check box disables the Signal threshold box, the Whole image, and VOI regions.

Example input image.
Output image, after N3 Correction.
Output inhomogeneity field.

Input Types

N3 Parameters

Example Usage Example input image. Output image, after N3 Correction. Output inhomogeneity field.

Example Usage

Example input image.

Output image, after N3 Correction.

Output inhomogeneity field.