neurostatx.statistics.harmonization

Functions

Name	Description
neuroCombat	Remove scanner/site effects from multi-site imaging data with ComBat.

neuroCombatfunction#

View source

def neuroCombat(dat, covars, batch_col, categorical_cols=None, continuous_cols=None, eb=True, parametric=True, mean_only=False, ref_batch=None)

Remove scanner/site effects from multi-site imaging data with ComBat.

Parameters

Name	Type	Default	Description
dat	DataFrame or np.array	required	Imaging features of shape (features, samples).
covars	DataFrame	required	Covariates including the batch/scanner column.
batch_col	str	required	Name of the batch (scanner) column in covars.
categorical_cols	list	None	Categorical covariate names to preserve. Defaults to None.
continuous_cols	list	None	Continuous covariate names to preserve. Defaults to None.
eb	bool	True	If True, use Empirical Bayes. Defaults to True.
parametric	bool	True	If True, use parametric adjustments. Defaults to True.
mean_only	bool	False	If True, adjust means only (no scaling). Defaults to False.
ref_batch	str	None	Batch used as the reference level. Defaults to None.

Returns

resultdict

Mapping with data (harmonized array), estimates, and info.

Examples

>>> import pandas as pd
>>> import numpy as np
>>> from neurostatx.statistics.harmonization import neuroCombat
>>> dat = np.random.RandomState(0).rand(4, 6)
>>> covars = pd.DataFrame({"scanner": [1, 1, 1, 2, 2, 2],
...                        "age": [20, 30, 40, 25, 35, 45]})
>>> out = neuroCombat(dat, covars, batch_col="scanner",
...                   continuous_cols=["age"])
>>> out["data"].shape
(4, 6)