Planned intervention: On Wednesday April 3rd 05:30 UTC Zenodo will be unavailable for up to 2-10 minutes to perform a storage cluster upgrade.
Published November 5, 2018 | Version 1.2.0
Dataset Open

High quality white matter reference tracts

  • 1. Division of Medical Image Computing, German Cancer Research Center

Description

Overview

This dataset contains segmentations of 72 white matter tracts obtained from 105 subjects included in the Human Connectome Project (HCP) young adult dataset (https://www.humanconnectome.org/study/hcp-young-adult). The folder names correspond to the ID of the HCP subjects. This dataset only contains the tracts. It does not contain the original DWI data. This has to be downloaded from the HCP website (it is free, but you have to register to get access).

The data is part of the following publication:  Wasserthal et al., TractSeg - Fast and accurate white matter bundle segmentation. NeuroImage (2018). If you use the data please cite the paper.

 

Details for generating corresponding whole brain tractograms

The tracts were extracted semi-automatically from whole-brain tractograms. For a detailed description of the tract segmentation process please refer to the paper. The following MRtrix (http://www.mrtrix.org/) commands were used to obtain the whole-brain tractograms:

5ttgen fsl T1w_acpc_dc_restore_brain.nii.gz 5TT.mif -premasked
dwi2response msmt_5tt Diffusion.nii.gz 5TT.mif RF_WM.txt RF_GM.txt RF_CSF.txt -voxels RF_voxels.mif -fslgrad Diffusion.bvecs Diffusion.bvals
dwi2fod msmt_csd Diffusion.nii.gz RF_WM.txt WM_FODs.mif RF_GM.txt GM.mif RF_CSF.txt CSF.mif -mask nodif_brain_mask.nii.gz -fslgrad Diffusion.bvecs Diffusion.bvals
tckgen -algorithm iFOD2 WM_FODs.mif output.tck -act 5TT.mif -backtrack -crop_at_gmwmi -seed_image nodif_brain_mask.nii.gz -maxlength 250 -minlength 40 -number 10M -cutoff 0.06 -maxnum 0

For "CA", "IFO_left", "IFO_right", "UF_left", "UF_right" we used tracking without anatomical constraints:

tckgen -algorithm iFOD2 WM_FODs.mif output.tck -seed_image nodif_brain_mask.nii.gz -maxlength 250 -minlength 40 -number 10M -cutoff 0.06 -maxnum 0

Due to their enormous size, the whole brain tractograms corresponding to the segmented tracts are not included this dataset. Please contact the author of the paper if you are interested in these tractograms.

 

Included tracts

1: AF_left         (Arcuate fascicle)
2: AF_right
3: ATR_left        (Anterior Thalamic Radiation)
4: ATR_right
5: CA              (Commissure Anterior)
6: CC_1            (Rostrum)
7: CC_2            (Genu)
8: CC_3            (Rostral body (Premotor))
9: CC_4            (Anterior midbody (Primary Motor))
10: CC_5           (Posterior midbody (Primary Somatosensory))
11: CC_6           (Isthmus)
12: CC_7           (Splenium)
13: CG_left        (Cingulum left)
14: CG_right   
15: CST_left       (Corticospinal tract
16: CST_right 
17: MLF_left       (Middle longitudinal fascicle)
18: MLF_right
19: FPT_left       (Fronto-pontine tract)
20: FPT_right 
21: FX_left        (Fornix)
22: FX_right
23: ICP_left       (Inferior cerebellar peduncle)
24: ICP_right 
25: IFO_left       (Inferior occipito-frontal fascicle) 
26: IFO_right
27: ILF_left       (Inferior longitudinal fascicle) 
28: ILF_right 
29: MCP            (Middle cerebellar peduncle)
30: OR_left        (Optic radiation) 
31: OR_right
32: POPT_left      (Parieto‐occipital pontine)
33: POPT_right 
34: SCP_left       (Superior cerebellar peduncle)
35: SCP_right 
36: SLF_I_left     (Superior longitudinal fascicle I)
37: SLF_I_right 
38: SLF_II_left    (Superior longitudinal fascicle II)
39: SLF_II_right
40: SLF_III_left   (Superior longitudinal fascicle III)
41: SLF_III_right 
42: STR_left       (Superior Thalamic Radiation)
43: STR_right 
44: UF_left        (Uncinate fascicle) 
45: UF_right 
46: CC             (Corpus Callosum - all)
47: T_PREF_left    (Thalamo-prefrontal)
48: T_PREF_right 
49: T_PREM_left    (Thalamo-premotor)
50: T_PREM_right 
51: T_PREC_left    (Thalamo-precentral)
52: T_PREC_right 
53: T_POSTC_left   (Thalamo-postcentral)
54: T_POSTC_right 
55: T_PAR_left     (Thalamo-parietal)
56: T_PAR_right 
57: T_OCC_left     (Thalamo-occipital)
58: T_OCC_right 
59: ST_FO_left     (Striato-fronto-orbital)
60: ST_FO_right 
61: ST_PREF_left   (Striato-prefrontal)
62: ST_PREF_right 
63: ST_PREM_left   (Striato-premotor)
64: ST_PREM_right 
65: ST_PREC_left   (Striato-precentral)
66: ST_PREC_right 
67: ST_POSTC_left  (Striato-postcentral)
68: ST_POSTC_right
69: ST_PAR_left    (Striato-parietal)
70: ST_PAR_right 
71: ST_OCC_left    (Striato-occipital)
72: ST_OCC_right
 

Cross-validation data splits

The following data splits were used for cross-validation in the TractSeg paper:

fold1 = ['992774', '991267', '987983', '984472', '983773', '979984', '978578', '965771', '965367', '959574', '958976', '957974', '951457', '932554', '930449', '922854', '917255', '912447', '910241', '907656', '904044']
fold2 = ['901442', '901139', '901038', '899885', '898176', '896879', '896778', '894673', '889579', '887373', '877269', '877168', '872764', '872158', '871964', '871762', '865363', '861456', '859671', '857263', '856766']
fold3 = ['849971', '845458', '837964', '837560', '833249', '833148', '826454', '826353', '816653', '814649', '802844', '792766', '792564', '789373', '786569', '784565', '782561', '779370', '771354', '770352', '765056']
fold4 = ['761957', '759869', '756055', '753251', '751348', '749361', '748662', '748258', '742549', '734045', '732243', '729557', '729254', '715647', '715041', '709551', '705341', '704238', '702133', '695768', '690152']
fold5 = ['687163', '685058', '683256', '680957', '679568', '677968', '673455', '672756', '665254', '654754', '645551', '644044', '638049', '627549', '623844', '622236', '620434', '613538', '601127', '599671', '599469']

Hyperparameters were optimized using fold 1-3 for training and fold 4 for validation.

The final 5-fold cross-validation (results reported in the TractSeg paper) was done by always training on 3 folds, selecting the best epoch by evaluating on the fourth fold and then reporting the final results (of the model from the best epoch) on the fifth fold.

The pretrained TractSeg model which will automatically be used when you download TractSeg was trained on fold1+fold2+fold3.

Please use the same data splits to make your work comparable.

 

Data format

From version 1.2.0 of this dataset onwards it uses the newest trackvis (trk) standard (using nibabel.streamlines API). Streamlines are saved in native voxel space and when loaded are transformed to coordinate space using the affine stored in the trk file header. In the previous versions of the dataset the older nibabel.trackvis API was used (streamlines are saved in real coordinate space and no affine is applied when loading them).

Files

HCP105_Zenodo_NewTrkFormat.zip

Files (44.0 GB)

Name Size Download all
md5:62db143099517a1a21299b380a751128
44.0 GB Preview Download

Additional details

Related works

Is supplement to
10.1016/j.neuroimage.2018.07.070 (DOI)

References