Warp_V-Norm

Description

This repository collects scripts to calculate the Warp V-Norm and the Warp V-Norm profile, as described in the paper "Supervised white matter bundle segmentation in glioma patients with transfer learning". The objective is to quantify structural alterations between two anatomically analogous fiber bundles by computing rigid, followed by a diffeomorphic registration between binar masks and mapping the resulting Warp V-Norm to the template bundle's skeleton profile.

This metric measures the structural differences between two fiber bundles assumed to be anatomically analogous. For the computation two NIfTI files representing the binary masks of these bundles (values 1 where the bundle exists, 0 otherwise) and a streamline file (.trk) representing the skeleton of the template (fixed) bundle are required.

The Warp V-Norm profile is computed through the following steps:

1. Registration: Estimates the optimal rigid and diffeomorphic transformation to align the moving bundle mask with the fixed bundle mask.
2. Warp V-Norm Calculation: Computes the norm-2 of the displacement vector field resulting from the total (rigid + diffeomorphic) transformation at each voxel. Higher values in the resulting Warp V-Norm map indicate greater structural divergence of the moving bundle from the fixed bundle in that spatial location.
3. Profile Mapping: Maps the computed Warp V-Norm map onto the points of the provided skeleton streamline (.trk file) of the fixed bundle. This generates a profile of the Warp V-Norm along the trajectory of the template bundle's skeleton, providing a one-dimensional representation of the structural alterations along the bundle's extent.

This profile allows for the visualization and quantification of how the moving bundle deviates structurally from the fixed bundle along its anatomical pathway.

This metric was designed to estimate the differences between:

• A subject's fiber bundle (considered as the moving bundle)
• And a healthy template bundle (considered the fixed bundle)

Important Notes:

• The skeleton provided should be defined along the trajectory of the fixed bundle.
• All files should be already aligned in the same space with a transformation based on corresponding anatomical images.

Usage

To compute Warp V Norm by computing the registration between the two bundles, and the norm-2 of the vectorial field of the estimated transformation in each voxel:

python compute_WarpVNorm.py <bundle_fixed> <bundle_moving>  <out_dir> 

To compute Warp V-Norm as in previous point, but additionally this command also maps Warp V-Norm values along the bundle-skeleton points:

python compute_WarpVNorm_profile_by_reg.py <bundle_fixed> <bundle_moving> <path_skeleton_trk> <n_points_profile> <out_dir> 

To compute Warp V Norm profile by using an already saved to file Warp V Norm map. In this case the registration is not computed since Warp V Norm was already estimated. The function simply maps the Warp V-Norm values along the skeleton points:

python compute_WarpVNorm_profile_by_nii.py <path_warp_vnorm_nii> <path_skeleton_trk> <n_points_profile> <path_warp_v_norm_profile_csv>