Groupwise Bundle Registration with DIPY

Groupwise Bundle Registration

When comparing white matter bundles across subjects, they first need to be coregistered to a common space. DIPY’s groupwise_slr function does this without requiring a pre-defined atlas: it iteratively aligns all bundles to an unbiased mean shape using Streamline Linear Registration (SLR).

What is SLR?

SLR (Streamline Linear Registration) optimizes a linear transform (translation, rotation, scaling, shearing) that minimizes the average distance between pairs of streamlines across two tractograms. The distance metric used is the minimum average direct-flip (MDF) distance, which is robust to streamline direction.

Loading bundles

The tutorial loads five left arcuate fasciculi from different subjects (or bundle samples):

from dipy.data import get_two_hemi_templates
from dipy.segment.bundles import groupwise_slr

# load 5 bundle instances as lists of streamline arrays
bundles = [...]  # list of StatefulTractogram objects

Running groupwise registration

from dipy.segment.bundles import groupwise_slr

bundles_reg, aff_list, d_list = groupwise_slr(
    bundles,
    nb_pts=20,
    verbose=True
)

The function returns:

• bundles_reg — the aligned bundle list
• aff_list — the per-subject linear transforms
• d_list — the pairwise distance at each iteration

The pairwise distance typically drops noticeably in the first few iterations and then plateaus, indicating convergence.

Before and after

The left panel shows five bundles before alignment; they occupy different parts of space. After groupwise_slr, all five overlap in a common coordinate frame, making cross-subject comparisons meaningful.