TractoInferno: A large-scale, open-source, multi-site database for machine learning dMRI tractography

# TractoInferno Machine Learning Tractography Dataset

The /derivatives folder contains the pre-split training/validation/testing datasets, each containing unique subjects with the following:

- T1W image
- DTI metrics maps (FA/AD/MD/RD)
- DWI image with bval/bvec
- fodf map + fodf peaks
- White matter/grey matter/csf masks
- DWI SH map (SH of order 6 fitted to the DWI signal, using the `descoteaux07` basis: https://dipy.org/documentation/1.3.0./theory/sh_basis/)
- Tractograms of the following delineated bundles
- AF_L
- AF_R
- CC_Fr_1
- CC_Fr_2
- CC_Oc
- CC_Pa
- CC_Pr_Po
- CG_L
- CG_R
- FAT_L
- FAT_R
- FPT_L
- FPT_R
- IFOF_L
- IFOF_R
- ILF_L
- ILF_R
- MCP
- MdLF_L
- MdLF_R
- OR_ML_L
- OR_ML_R
- POPT_L
- POPT_R
- PYT_L
- PYT_R
- SLF_L
- SLF_R
- UF_L
- UF_R

All tractograms contain compressed streamlines to reduce disk space, which means that the step size is variable. If your algorithm requires a fixed step size, you have to manually resample the streamlines, which can be done using SCILPY (https://github.com/scilus/scilpy) and the scil_resample_streamlines.py script: https://github.com/scilus/scilpy/blob/master/scripts/scil_resample_streamlines.py

To evaluate a candidate tractogram, refer to: https://github.com/scil-vital/TractoInferno/