A valuable contribution, which spares us from implementing the N3 method. It has basic documentation and a very complex example, but it lacks a simplest example.

Source code provided, as well as some images.

It compiles after ITK was configured to use reivew code. The complex example runs noticeably faster than the code I used:

typedef itk::N4MRIBiasFieldCorrectionImageFilter<RealImageType, CharImageType> CorrecterType;

CorrecterType::Pointer correcter=CorrecterType::New();

correcter->SetInput(reader->GetOutput());

correcter->Update();

Code is usable, but the examples provided are long, and a straightforward use of the filter takes much longer than the provided examples.

I did not check the included data, I used my own data.

Everyone who uses MR images.

The authors present an implementation of a published method for MRI bias correction. The code follows ITK style, is accompanied by guidelines on parameter initialization and references to the literature. The authors supply the code with the data they used to test the code.

n/a

n/a

The code uses ITK, the authors do provide images, and sufficient details to replicate the results.

The code is compiled with review enabled. We were able to reproduce the results authors present, and successfully apply the implementation to our data.

The code promotes best practices of open source and ITK programming.

The code is usable, the authors provide flexibility to work with either 2d or 3d images, the code includes not only filters, but also the test driver program to simplify usage.

The coding style seems to be very good, the code is readable and follows ITK design. I did not observe any features that would make this code non-portable, but personally have compiled it only on linux.

The authors provide reasonable testing data sets.

CORRECTION: the accompanying document does not stress the meaning of the shrink factor. My original experiments we using shrink factor 1, which results in no resampling. With the shrink factor 4, the code takes under 1 minute for the same data. At the same time, I cannot distinguish visually the bias field from the one produced with shrink factor 1. I withdraw my complaint about high execution time.

The code is rather slow on 3d data: 256x256x124 image takes about 50 min to process. The code is not multi-threaded.

Mask generation can be challenging due to the nature of input data -- field inhomogeneities may not (and do not, in our case) allow to generate the mask with simple Otsu threshold.

The algorithm requires setting of a relatively large number of parameters, and due to significant execution time, parameter tuning is not easy. We were able to achieve visually pleasing results by increasing the m_BiasFieldFullWidthAtHalfMaximum parameter from its default value.

A wrapper module for 3D Slicer (http://www.slicer.org) has been implemented by Nicolas Rannou (Brigham and Women's Hospital), and can be downloaded from NAMIC SandBox here: http://svn.na-mic.org/NAMICSandBox/trunk/MRIBiasCorrectionFilter/. This module has preview mode, which allows to apply correction to a single slice, making parameter selection easier. The use of Slicer allows to use additional filters to construct and edit mask that can be used in bias correction.

I am using ITK 3.12, and I get compilation errors. What am I missing?