Python implementation of Wu et al (2018)'s registration fusion

Overview

reg-fusion

logo
Projection of a central sulcus probability map using the RF-ANTs approach (right hemisphere shown).

This is a Python implementation of Wu et al (2018)'s registration fusion methods to project MRI data from standard volumetric coordinates, either MNI152 or Colin27, to Freesurfer's fsaverage. This tool already available in the original MATLAB-based version provided by Wu et al, which works well out of the box. However, given Python's increasing stake in neuroimaging analysis, a pure Python version may be useful.

A huge thank you to Wu et al for making their excellent tool openly available! If you use this package, please cite the original:

Wu J, Ngo GH, Greve DN, Li J, He T, Fischl B, Eickhoff SB, Yeo BTT. Accurate nonlinear mapping between MNI volumetric and FreeSurfer surface coordinate systems, Human Brain Mapping 39:3793–3808, 2018.

Installation

This package requires Python 3. Installing regfusion is simple with pip:

pip install regfusion

If you want to build regfusion directly from source code, use the following code:

git clone https://github.com/danjgale/reg-fusion
cd reg-fusion
python setup.py install

Command-line interface

Registration fusion can be ran on the command-line using regfusion. The flags correspond to the original implemenation, with the exception of -t, which is specific to regfusion (see Notes).

usage: regfusion [-h] [-s input_vol] [-o output_dir] [-p template_type] [-r RF_type] [-i interp] [-t out_type]

optional arguments:
  -h, --help        show this help message and exit
  -s input_vol      Absolute path to input volume. Input should be in nifti format
  -o output_dir     Absolute path to output directory
  -p template_type  Type of volumetric template used in index files. Use MNI152_orig or Colin27_orig when -r is RF_ANTs. Use MNI152_norm or Colin27_norm when
                    -r is RF_M3Z. Otherwise, an exception is raised. Ensure that the template matches the standard space of -i (i.e., use MNI152_* if -i is
                    in MNI152-space). Default: MNI152_orig
  -r RF_type        Type of Registration Fusion approaches used to generate the mappings (RF_M3Z or RF_ANTs). RF_M3Z is recommended if data was registered
                    from subject's space to the volumetric atlas space using FreeSurfer. RF_ANTs is recommended if such registrations were carried out using
                    other tools, especially ANTs. Default: RF_ANTs
  -i interp         Interpolation (linear or nearest). If -g is label.gii, then interpolation is always set to nearest and a warning is raised. Default:
                    linear
  -t out_type       File type of surface files. nii.gz is true to the original Wu et al (2018) implementation. Note that gifti formats, either func.gii or
                    label.gii, are often preferred. Default: nii.gz

Python API

The CLI simply calls the main underlying function, vol_to_fsaverage. This function can imported directly in Python. In addition to saving the files to out_dir, the absolute file paths of the left and right surface files are returned.

vol_to_fsaverage(input_img, out_dir, template_type='MNI152_orig', 
                 rf_type='RF_ANTs', interp='linear', out_type='nii.gz'):

    Project volumetric data in standard space (MNI152 or Colin27) to 
    fsaverage 

    Parameters
    ----------
    input_img : niimg-like
        Input image in standard space (i.e. MNI152 or Colin27)
    out_dir : str
        Path to output directory (does not need to already exist)
    template_type : {'MNI152_orig', 'Colin27_orig', 'MNI152_norm', 'Colin27_norm'}
        Type of volumetric template used in index files. Use 'MNI152_orig' or 
        'Colin27_orig' when `rf_type` is 'RF_ANTs'. Use 'MNI152_norm' or 
        'Colin27_norm' when `rf_type` is 'RF_M3Z'. Otherwise, an exception is 
        raised. Ensure that the template matches the standard space of 
        `input_img` (i.e., use MNI152_* if `input_img` is in MNI152-space). By 
        default 'MNI152_orig'.
    rf_type : {'RF_ANTs', 'RF_M3Z'}
        Type of Registration Fusion approaches used to generate the mappings.
        RF-M3Z is recommended if data was registered from subject's space to 
        the volumetric atlas space using FreeSurfer. RF-ANTs is recommended if 
        such registrations were carried out using other tools, especially 
        ANTs. By default 'RF_ANTs'
    interp : {'linear', 'nearest'}, optional
        Interpolation approach. If `out_type` is 'label.gii', then interpolation 
        is always set to 'nearest'. By default 'linear'
    out_type : {'nii.gz, 'func.gii', 'label.gii'}, optional
        File type of surface files. Default is 'nii.gz', which is true to the 
        original Wu et al (2018) implementation. Note that gifti 
        formats, either 'func.gii' or 'label.gii', are often preferred.

    Returns
    ----------
    str, str
        Absolute paths to left and right hemisphere output files, respectively

Examples

1. MNI to fsaverage (default)

For example, the default RF-ANTs implementation (preferred) with MNI data would be:

CLI:
regfusion -s mni_input.nii.gz -o output

Python:
from regfusion import vol_to_fsaverage
lh, rh = vol_to_fsaverage('mni_input.nii.gz', 'output')

True to the original implementation, two surface files (one each hemisphere) are saved to the output directory with the RF method and template embedded in the file names:

output/
  lh.mni_input.allSub_RF_ANTs_MNI152_orig_to_fsaverage.nii.gz
  rh.mni_input.allSub_RF_ANTs_MNI152_orig_to_fsaverage.nii.gz

2. MNI to fsaverage (GIfTI)

It may be preferred to generate GIfTI files instead of the default NIfTI:

CLI:
regfusion -s mni_input.nii.gz -o output -t func.gii

Python:
from regfusion import vol_to_fsaverage
lh, rh = vol_to_fsaverage('mni_input.nii.gz', 'output', out_type='func.gii')

The output, which will have the appropriate GIfTI file extensions:

output/
  lh.mni_input.allSub_RF_ANTs_MNI152_orig_to_fsaverage.func.gii
  rh.mni_input.allSub_RF_ANTs_MNI152_orig_to_fsaverage.func.gii

3. Projecting to label.gii

Should you wish to project a binary mask (e.g., to display a region of interest), you may consider setting the output type, -t, to label.gii. In this case, interpolation, -i, will always be set to nearest to retain the original voxel values/labels. If not explicitly set with -i, interpolation will be overwritten to nearest and warning is raised.

For example:

CLI:
regfusion -s mni_input.nii.gz -o output -i nearest -t label.gii

Python:
from regfusion import vol_to_fsaverage
lh, rh = vol_to_fsaverage('mni_input.nii.gz', 'output', interp='nearest', out_type='label.gii')

The output, which will have the appropriate GIfTI file extensions:

output/
  lh.mni_input.allSub_RF_ANTs_MNI152_orig_to_fsaverage.label.gii
  rh.mni_input.allSub_RF_ANTs_MNI152_orig_to_fsaverage.label.gii

4. MNI to fsaverage with RF-M3Z

And finally, the RF-M3Z method can be used if that is preferred:

CLI:
regfusion -i mni_input.nii.gz -o output -p MNI152_norm -r RF_M3Z

Python:
from regfusion import vol_to_fsaverage
lh, rh = vol_to_fsaverage('mni_input.nii.gz', 'output', template_type='MNI152_norm', rf_type='RF_M3Z')

The output, with different file names reflecting the method/template used:

output/
  lh.mni_input.allSub_RF_M3Z_MNI152_norm_to_fsaverage.nii.gz
  rh.mni_input.allSub_RF_M3Z_MNI152_norm_to_fsaverage.nii.gz

Notes

regfusion implements the same two registration fusion approaches by Wu et al, and is validated against the original MATLAB version (see tests/). However, there are some differences in the API:

  • regfusion does not have the -n flag that determines the number of subjects used to create the average mapping. That is because the standalone scripts of the MATLAB versions only uses all 1490 subjects, and thus regfusion does too
  • regfusion does not have the -m flag because no MATLAB is required
  • regfusion does not have the -f flag because, technically, Freesurfer is not required. However, it is strongly recommended that you have a freely available Freesurfer license because we are ultimately projecting to Freesurfer's fsaverage
  • Unlike the original MATLAB version, regfusion has a -t flag (out_type in vol_to_fsaverage; see above for description). The original MATLAB version outputs NIfTI images (regfusion default), but this option lets regfusion output to GIfTIs, which are generally preferred for surface files. Users are encouraged to set -t/out_type to one of the GIfTI output types if they find that GIfTIs are more suitable for their needs

Some useful things to know:

  • Wu et al show that RF-ANTs is generally the better approaches of the two, which is why it's the default in regfusion. RF-M3Z seems best-suited if the normalization was performed via Freesurfer.
  • As Wu et al emphasize, the actual best practice here avoid projecting standard volumetric coordinates (e.g., MNI) to fsaverage altogether. Alternatives include performing all you analyses in subject/native volumetric coordinates and projecting that data to fsaverage, based on Freesurfer's recon-all. Or, perform analyses directly in fsaverage after running recon-all. Projecting data from one standard coordinates space to another is loses precision at each step (see Wu et al for details). Neverthless, people do this all the time and these registration fusion approaches ensure that these projections are as accurate as possible.
  • Relating to the previous point: If you do project from MNI/Colin coordinates to fsaverage, it's probably a wise idea to find a way to still show your data in volume-space too (e.g., as supplementary figures/material).

References

Wu J, Ngo GH, Greve DN, Li J, He T, Fischl B, Eickhoff SB, Yeo BTT. Accurate nonlinear mapping between MNI volumetric and FreeSurfer surface coordinate systems, Human Brain Mapping 39:3793–3808, 2018.

Owner
Dan Gale
Neuroscience PhD candidate with an interest in data science and software development.
Dan Gale
Position detection system of mobile robot in the warehouse enviroment

Autonomous-Forklift-System About | GUI | Tests | Starting | License | Author | 🎯 About An application that run the autonomous forklift paletization a

Kamil GoÅ› 1 Nov 24, 2021
a basic code repository for basic task in CV(classification,detection,segmentation)

basic_cv a basic code repository for basic task in CV(classification,detection,segmentation,tracking) classification generate dataset train predict de

1 Oct 15, 2021
Uncertainty-aware Semantic Segmentation of LiDAR Point Clouds for Autonomous Driving

SalsaNext: Fast, Uncertainty-aware Semantic Segmentation of LiDAR Point Clouds for Autonomous Driving Abstract In this paper, we introduce SalsaNext f

308 Jan 04, 2023
Hcaptcha-challenger - Gracefully face hCaptcha challenge with Yolov5(ONNX) embedded solution

hCaptcha Challenger 🚀 Gracefully face hCaptcha challenge with Yolov5(ONNX) embe

593 Jan 03, 2023
A python script to convert images to animated sus among us crewmate twerk jifs as seen on r/196

img_sussifier A python script to convert images to animated sus among us crewmate twerk jifs as seen on r/196 Examples How to use install python pip i

41 Sep 30, 2022
A minimal implementation of face-detection models using flask, gunicorn, nginx, docker, and docker-compose

Face-Detection-flask-gunicorn-nginx-docker This is a simple implementation of dockerized face-detection restful-API implemented with flask, Nginx, and

Pooya-Mohammadi 30 Dec 17, 2022
NER for Indian languages

CL-NERIL: A Cross-Lingual Model for NER in Indian Languages Code for the paper - https://arxiv.org/abs/2111.11815 Setup Setup a virtual environment Th

Akshara P 0 Nov 24, 2021
Repo for parser tensorflow(.pb) and tflite(.tflite)

tfmodel_parser .pb file is the format of tensorflow model .tflite file is the format of tflite model, which usually used in mobile devices before star

1 Dec 23, 2021
MILK: Machine Learning Toolkit

MILK: MACHINE LEARNING TOOLKIT Machine Learning in Python Milk is a machine learning toolkit in Python. Its focus is on supervised classification with

Luis Pedro Coelho 610 Dec 14, 2022
SwinIR: Image Restoration Using Swin Transformer

SwinIR: Image Restoration Using Swin Transformer This repository is the official PyTorch implementation of SwinIR: Image Restoration Using Shifted Win

Jingyun Liang 2.4k Jan 08, 2023
A TensorFlow Implementation of "Deep Multi-Scale Video Prediction Beyond Mean Square Error" by Mathieu, Couprie & LeCun.

Adversarial Video Generation This project implements a generative adversarial network to predict future frames of video, as detailed in "Deep Multi-Sc

Matt Cooper 704 Nov 26, 2022
Official implementation of the paper Do pedestrians pay attention? Eye contact detection for autonomous driving

Do pedestrians pay attention? Eye contact detection for autonomous driving Official implementation of the paper Do pedestrians pay attention? Eye cont

VITA lab at EPFL 26 Nov 02, 2022
Block-wisely Supervised Neural Architecture Search with Knowledge Distillation (CVPR 2020)

DNA This repository provides the code of our paper: Blockwisely Supervised Neural Architecture Search with Knowledge Distillation. Illustration of DNA

Changlin Li 215 Dec 19, 2022
Ranger deep learning optimizer rewrite to use newest components

Ranger21 - integrating the latest deep learning components into a single optimizer Ranger deep learning optimizer rewrite to use newest components Ran

Less Wright 266 Dec 28, 2022
NFT-Price-Prediction-CNN - Using visual feature extraction, prices of NFTs are predicted via CNN (Alexnet and Resnet) architectures.

NFT-Price-Prediction-CNN - Using visual feature extraction, prices of NFTs are predicted via CNN (Alexnet and Resnet) architectures.

5 Nov 03, 2022
Unofficial implementation (replicates paper results!) of MINER: Multiscale Implicit Neural Representations in pytorch-lightning

MINER_pl Unofficial implementation of MINER: Multiscale Implicit Neural Representations in pytorch-lightning. 📖 Ref readings Laplacian pyramid explan

AI葵 51 Nov 28, 2022
UnFlow: Unsupervised Learning of Optical Flow with a Bidirectional Census Loss

UnFlow: Unsupervised Learning of Optical Flow with a Bidirectional Census Loss This repository contains the TensorFlow implementation of the paper UnF

Simon Meister 270 Nov 06, 2022
Implementation of RegretNet with Pytorch

Dependencies are Python 3, a recent PyTorch, numpy/scipy, tqdm, future and tensorboard. Plotting with Matplotlib. Implementation of the neural network

Horris zhGu 1 Nov 05, 2021
PyTorch implementation of SimSiam: Exploring Simple Siamese Representation Learning

SimSiam: Exploring Simple Siamese Representation Learning This is a PyTorch implementation of the SimSiam paper: @Article{chen2020simsiam, author =

Facebook Research 834 Dec 30, 2022
Author Disambiguation using Knowledge Graph Embeddings with Literals

Author Name Disambiguation with Knowledge Graph Embeddings using Literals This is the repository for the master thesis project on Knowledge Graph Embe

12 Oct 19, 2022