Anatomy-to-Tract Mapping Infers White Matter Pathways Without Diffusion Streamline Propagation

This repository contains the code and pretrained model weights for the paper:

Environment

The code has been tested and successfully executed with the following environment:

• Python: 3.10.10

• PyTorch: 2.5.1

• MRtrix3

• Scilpy

• ANTs

All required Python libraries are listed in requirements.txt.

Note: MRtrix3, ANTs, and Scilpy are external tools and must be installed separately.

Input Directory Structure

To run the code successfully, place the input data inside a folder named data, and organize each subject’s directory as follows:

data/
└── sub-0000/
    ├── anat/
    │   └── sub-0000__T1w.nii.gz
    ├── mask/
    │   ├── sub-0000__mask_gm.nii.gz
    │   └── sub-0000__mask_wm.nii.gz
    ├── surf/
    │   ├── lh.white.surf.gii
    │   ├── rh.white.surf.gii
    │   ├── lh.white.normals.func.gii
    │   └── rh.white.normals.func.gii
    └── mri/
        └── brainmask.nii

Preprocessing Recommendation

We recommend preprocessing the T1wimage using FreeSurfer’s recon-all pipeline prior to running ATM. Required inputs—such as cortical surfaces, gray and white matter masks, and the brain mask—can be extracted from the FreeSurfer output as needed.

Command

To run the inference, execute:

python infer.py --bundle all --sub_id sub-1135 --num_streamline 3000