Skip to content

Harmonization of multi-site imaging data with ComBat (Python)

License

Notifications You must be signed in to change notification settings

Jfortin1/neuroCombat

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

38 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Multi-site harmonization in Python with neuroCombat

License: MIT Version PythonVersion

This is the maintained and official version of neuroCombat (previously hosted here) introduced in our our recent paper.

Installation

neuroCombat is hosted on PyPI, and the easiest way to install neuroCombat is to use the pip command:

pip install neuroCombat

Usage

The neuroCombat function performs harmonization

from neuroCombat import neuroCombat
import pandas as pd
import numpy as np

# Getting example data
# 200 rows (features) and 10 columns (scans)
data = np.genfromtxt('testdata/testdata.csv', delimiter=",", skip_header=1)

# Specifying the batch (scanner variable) as well as a biological covariate to preserve:
covars = {'batch':[1,1,1,1,1,2,2,2,2,2],
          'gender':[1,2,1,2,1,2,1,2,1,2]} 
covars = pd.DataFrame(covars)  

# To specify names of the variables that are categorical:
categorical_cols = ['gender']

# To specify the name of the variable that encodes for the scanner/batch covariate:
batch_col = 'batch'

#Harmonization step:
data_combat = neuroCombat(dat=data,
    covars=covars,
    batch_col=batch_col,
    categorical_cols=categorical_cols)["data"]

Optional arguments

  • eb : True or False. Should Empirical Bayes be performed? If False, the harmonization model will be fit for each feature separately. This is equivalent to performing a location/shift (L/S) correction to each feature separately (no information pooling across features).

  • parametric : True or False. Should parametric adjustements be performed? True by default.

  • mean_only : True or False. Should only be means adjusted (no scaling)? False by default

  • ref_batch : batch name to be used as the reference batch for harmonization. None by default, in which case the average across scans/images/sites is taken as the reference batch.

Output

Since version 0.2.10, the neuroCombat function outputs a dictionary with 3 elements:

  • data: A numpy array of the harmonized data, with the same dimension (shape) as the input data.
  • estimates: A dictionary of the neuroCombat estimates; useful for visualization and understand scanner effects.
  • info: A dictionary of the inputs needed for ComBat harmonization (batch/scanner information, etc.)

To simply return the harmonized data, one can use the following:

data_combat = neuroCombat(dat=dat, ...)["data"]

where ... are the user-specified arguments needed for harmonization.