When the images are noisy, MRI data processing steps such as coregistation or segmentation are more troublesome. A common method to suppress noise in images is to apply Gaussian smoothing. However, Gaussian smoothing blurs edges in the images, causing mixing of measurements across different tissues. This is a problem because the edges are important sources of information for coregistration and segmentation algorithms. A solution for this problem is using more sophisticated smoothing/denoising algorithms that respect the edges.

An impressive edge preserving image smoothing method is non-linear anisotopic diffusion based smoothing*. Currently there are two different flavors implemented in Segmentator**:

```
segmentator_filters /path/to/image.nii.gz --nr_iterations 10
```

<img src="pictures/nonlinear_anisotropic_filter_01.gif" width="700">

___
There is also a less conservative version which smooths out the noise more aggressively:
```
segmentator_filters /path/to/image.nii.gz --smoothing CURED --nr_iterations 30 --save_every 1
```
<img src="pictures/nonlinear_anisotropic_filter_02.gif" width="300">

### Notes
* *: Non-linear anisotopic diffusion based smoothing refers to the definition of Weickert (1998) here. This should not be confused with what Perona & Malik (1990) referred to as non-linear anisotropic smoothing. In Weickert's terminology, the method used by Perona & Malik is called non-linear _isotropic_ smoothing.
* **: This implementation has high RAM usage. For reference, an image with 384×384×256 voxels uses around 8 GB of memory. You can crop or mask your images for less memory consumption and faster processing.

### Alternative implementation

- I have become aware that [AFNI](https://afni.nimh.nih.gov/) has a much older implementation of this algorithm ([`3danisosmooth`](https://afni.nimh.nih.gov/pub/dist/doc/program_help/3danisosmooth.html)). Do consider using it as an alternative, as it has a few different knobs and dials to control it compared to my implementation here.

### References
* Weickert, J. (1998). Anisotropic diffusion in image processing. ECMI.

* Mirebeau, J.-M., Fehrenbach, J., Risser, L., & Tobji, S. (2015). Anisotropic Diffusion in ITK, 1–9. <http://arxiv.org/abs/1503.00992>

* Perona, P., & Malik, J. (1990). Scale-space and edge detection using anisotropic diffusion. IEEE Transactions on Pattern Analysis and Machine Intelligence, 12(7), 629–639. <http://doi.org/10.1109/34.56205>