Data¶

In order to participate you should first register via "Join" button and after approved you will have the links for downloading the data in "Dataset" page.

The dataset comprise 75 clinical cases with T1w Structural and Diffusion Weighted (DW) modalities.
The scans were acquired on a 3T Signa machine (GE healthcare, Milwaukee). The DW protocol had 64
gradient directions at B0 = 1000 and one scan at B0 = 0. T1w with no contrast injection was acquired at
1x1x1 mm^3 resolution and the DW with 1x1x2.6 mm^3. DW was processed with MrDiffusion toolbox [1]
for Matlab and registered to the high resolution T1w. Finally T1w scans were skull stripped with BET [2] and
resliced to 1.25x1.25x1.25 mm^3, DW scans (registered to T1w) were resliced to 1.25x1.25x1.25 mm^3 as well.
Each T1w and DW scan is stored in a separate nifti file with 128x144x128 spatial dimension.

White matter tracts mapping annotations were acquired in a semi-manual process for each scan. Optic Radiation (OR) and Corticospinal tracts (CST) (left and right) were chosen as they are of significant interest during neuro-surgical planning.

Tracts annotations are provided only for 60 cases (train set) while the rest 15 cases are used for participants
algorithms evaluation, so only the input cases scans are provided (T1w and DW).

NOTE: for 4 out of the 60 train cases there are no CST tracts annotations, due to difficulties in tractography reconstruction.

Submission¶

Each participant will have to segment the 2 required types of tracts (left and right side for each) for the 15 test cases provided. Each tract segmentation should be in the same format as the tracts provided for training: binary segmentation map 128x144x128 with the same affine as provided in T1w/DW of specific case,  each voxel containing the desired tract should contain an integer > 0 or 0 otherwise. For each tract there should be stored in a separate nifti file (in the same way it is provided for the training cases), so that the overall submitted files should be as follows:

submission_name/
├── case_61/
│   ├── CST_left.nii.gz
│   ├── CST_right.nii.gz
│   ├── OR_left.nii.gz
│   └── OR_right.nii.gz
├── case_62/
│   ├── CST_left.nii.gz
│   ├── CST_right.nii.gz
│   ├── OR_left.nii.gz
│   └── OR_right.nii.gz
├──

For each test case there should be a separate folder: case_61, case_62, etc. All cases folders should be zipped to a single file having a name that will later on appear in participants leaderboard table.

Evaluation¶

Submissions will be evaluated with a Dice [3] and Jaccard [4] scores.

For each type of tract (OR right, OR left, CST right, CST left) an average score over 15 test cases will be taken and used for ranking the participants algorithms in the leaderboard.

References¶

[1] https://web.stanford.edu/group/vista/cgi-bin/wiki/index.php/MrDiffusion

[2] M. Jenkinson, M. Pechaud, and S. Smith. BET2: MR-based estimation of brain, skull and scalp surfaces. In Eleventh Annual Meeting of the Organization for Human Brain Mapping, 2005.

[3] L.R. Dice, “Measures of the amount of ecologic association between
species,” Ecology, vol. 26, no. 3, pp. 297-302, 1945.

[4] L. Hamers, “Similarity measures in scientometric research: The Jaccard index versus Salton's
cosine formula,” Information Processing and Management, vol. 25, no. 3, pp. 315-18, 1989