dipy_denoise

Synopsis

Denoise diffusion data, selecting the method from the data.

With method='auto', the method is chosen per input file from the number of volumes, the number of diffusion-weighted volumes and an estimate of their SNR. See [1], [2] and [3] for the underlying methods.

An explicit method takes precedence over the selection thresholds: min_volumes, min_directions and min_snr are only read with method='auto'. In that case the criteria are checked in the following order and the first one that matches decides:

1. fewer than min_volumes volumes: ‘nlmeans’;

2. fewer than min_directions diffusion-weighted volumes: ‘mppca’;

3. estimated SNR below min_snr: ‘mppca’;

4. otherwise ‘patch2self’, or ‘mppca’ when no bvalues_files is given.

Usage

dipy_denoise [OPTIONS] input_files

Input Parameters

• input_files

  Path to the input volumes. This path may contain wildcards to process multiple inputs at once.

General Options

• --bvalues_files

  Path to the b-values files, one per input volume. Required for ‘patch2self’ and used by ‘auto’ to separate b0 volumes from diffusion-weighted ones. Without it, ‘auto’ treats every volume as diffusion-weighted. (default: None)

• --method

  Denoising method: ‘auto’, ‘patch2self’, ‘mppca’ or ‘nlmeans’. Any value other than ‘auto’ forces that method and the min_volumes, min_directions and min_snr thresholds are ignored. (default: auto)

• --b0_threshold

  Threshold used to find b0 volumes. (default: 50)

• --min_volumes

  ‘auto’ only. Below this number of volumes, ‘nlmeans’ is used. Checked first: it takes precedence over min_directions and min_snr. (default: 10)

• --min_directions

  ‘auto’ only. Minimum number of diffusion-weighted volumes for ‘patch2self’; otherwise ‘mppca’ is used. Checked after min_volumes and before min_snr. (default: 30)

• --min_snr

  ‘auto’ only. Minimum estimated SNR of the diffusion-weighted volumes for ‘patch2self’; otherwise ‘mppca’ is used. Checked last, only when min_volumes and min_directions are both satisfied. (default: 5.0)

• --patch_radius

  Radius of the local patch for ‘mppca’ and ‘nlmeans’. (default: 2)

• --model

  Linear model for ‘patch2self’: ‘ols’, ‘ridge’ or ‘lasso’. (default: ols)

• --ver

  Version of the Patch2Self algorithm, 1 or 3. (default: 3)

• --clip_negative_vals

  ‘patch2self’ only. Set negative values after denoising to 0. (default: False)

• --sigma

  ‘nlmeans’ only. Noise standard deviation; 0 estimates it from the data. (default: 0.0)

• --block_radius

  ‘nlmeans’ only. Block size is 2 x block_radius + 1. (default: 5)

• --rician

  ‘nlmeans’ only. Assume Rician rather than Gaussian noise. (default: True)

Output Options

• --out_dir

  Output directory. (default: current directory)

• --out_denoised

  Name of the resulting denoised volume. (default: dwi_denoised.nii.gz)

References

[1]

Shreyas Fadnavis, Joshua Batson, and Eleftherios Garyfallidis. Patch2Self: Denoising Diffusion MRI with Self-Supervised Learning. In Hugo Larochelle, Marc’Aurelio Ranzato, Raia Hadsell, Maria-Florina Balcan, and Hsuan-Tien Lin, editors, Advances in Neural Information Processing Systems, volume 33, 16293–16303. Curran Associates, Inc., 2020. URL: https://proceedings.neurips.cc/paper_files/paper/2020/file/bc047286b224b7bfa73d4cb02de1238d-Paper.pdf.

[2]

Jelle Veraart, Els Fieremans, Ileana O. Jelescu, Florian Knoll, and Dmitry S. Novikov. Gibbs ringing in diffusion MRI. Magnetic Resonance in Medicine, 76(1):301–314, July 2016. URL: https://doi.org/10.1002/mrm.25866, doi:10.1002/mrm.25866.

[3]

Maxime Descoteaux, Nicolas Wiest-Daesslé, Sylvain Prima, Christian Barillot, and Rachid Deriche. Impact of Rician adapted Non-Local Means filtering on HARDI. In Dimitris Metaxas, Leon Axel, Gabor Fichtinger, and Gábor Székely, editors, Medical Image Computing and Computer-Assisted Intervention (MICCAI), volume 11, 122–130. Berlin, Heidelberg, 2008. Springer Berlin Heidelberg. doi:10.1007/978-3-540-85990-1_15.

Garyfallidis, E., M. Brett, B. Amirbekian, A. Rokem, S. Van Der Walt, M. Descoteaux, and I. Nimmo-Smith. Dipy, a library for the analysis of diffusion MRI data. Frontiers in Neuroinformatics, 1-18, 2014.