Skip to content

Quickstart

A five-minute lesson: install, normalize one T1-w image, check the result.

Install

pip install intensity-normalization  # or: uv add intensity-normalization

Get a brain mask

Most methods expect a brain mask — a rough one is fine, it only needs to remove most non-brain tissue. ROBEX, HD-BET, and SynthStrip all work. If the image is already skull-stripped, skip the mask: the foreground is estimated as the positive voxels.

Normalize one image

FCM normalization is a good default for T1-w brain images: it scales the white-matter mean to 1.

intensity-normalize fcm t1w.nii.gz -m brain_mask.nii.gz -o t1w_norm.nii.gz

import nibabel as nib
import intensity_normalization as inorm

image = nib.load("t1w.nii.gz")
mask = nib.load("brain_mask.nii.gz")

normed = inorm.fcm(image, mask=mask)
nib.save(normed, "t1w_norm.nii.gz")

The output has the same type as the input: numpy in → numpy out, nibabel in → nibabel out, affine and header preserved.

Check the result

Plot the foreground histogram before and after — the fastest way to catch a bad mask. Install the plotting extra and rerun with -p:

pip install "intensity-normalization[plot]"  # or: uv add ...
intensity-normalize fcm t1w.nii.gz -m brain_mask.nii.gz -p

After FCM normalization of T1-w images, the white-matter peaks should sit at the norm value (1 by default).

Where next