Sub-Challenge #1

Challenge name

The signal forecast: generalizability of diffusion signal representations

Purpose and relevance of the challenge

With this challenge we aim to understand the current ability of the field at describing the signal measured in diffusion MRI. The challenge consists of a number of signals sampled from datasets acquired in human and mice with different diffusion sequences. Participants will be provided with a subsampled set of the acquired data and are asked to predict the remaining – unseen – data. The acquired data includes different types of acquisition strategies, such as multi-shell and DSI-like pulsed gradient spin-echo (PGSE) [1], as well as double diffusion encoding (DDE) [2] and double oscillating diffusion encoding (DODE) [3], suitable for a large number of fit approaches. The outcome of this challenge will allow to objectively evaluate the generalizability and appropriateness of current techniques.



The challenge includes in-vivo brain data sampled with PGSE in a human volunteer, and with DDE / DODE in mice, ex-vivo. You can choose which acquisition you would like to predict, and can choose any or all acquisitions strategies.

The two provided datasets are composed as follows:

  • PGSE. A subset of the data measured in five voxels, representative of different tissue configurations, extracted from the MASSIVE dataset4 (3820 unique diffusion-weighted volumes). The signals measured in each voxel include 2580 unique data points acquired with a multi-shell strategy, and 1240 data points acquired with a DSI-like strategy. The two acquisitions were collected in 19 separate sessions. The multi-shell and the DSI-like sequence were NOT performed with identical imaging parameters, but with identical diffusion gradients settings. Participants can choose whether they prefer to work with the multi-shell or the DSI-like acquisition.
  • DDE and DODE. A subset of the data from five voxels acquired in a mouse brain, representative of different tissue configurations will be provided. The full dataset consists of DDE with 2 different diffusion times and DODE with 5 different frequencies, with 5 b- values and 72 directions each (2520 diffusion weighted volumes in total).

Link to Data : See Registration and Data Access page

Participation (Data given to the participants)

Participants will be provided with part of the measured signals for 5 different tissue configurations. Additionally, they will receive complete details of the acquisition protocol (imaging parameters, diffusion gradients properties) corresponding to the measured data, including both provided and unseen signals. Regarding multi-shell PGSE, we will provide, for each signal, 495 of the 2580 diffusion-weighted measurements, as well as 20 b = 0s/mm2. Regarding DSI-like PGSE, we provide 480 of the 1240 measured points and 20 b = 0s/mm2. The signals will be distributed in text format in separate files for each acquisition strategy, and do not correspond to the same voxels. For the DDE/DODE data we will provide data from 1 DDE and 3 DODE waveforms with 4 b-values and 72 directions each (1152 / 2520 measurements). Each text file will contain the signal in floating format, with each column representing a different signal, and each row a measurement. The acquisitions details of each measurement can be found in a companion text file “*_AcqParams.txt”. Sample scripts to load the data and the acquisitions details will be provided for popular working environments (MATLAB, Python, C/C++).


Participants are asked to submit signal predictions of the unseen data for one of the provided datasets. We ask to submit a zip file containing a textual description of the performed analysis (info.txt) together with the signal predictions (submission.txt).

Please include the following details in the info.txt:

  1. the submission name and abbreviation;
  2. the team name;
  3. team members who made meaningful contributions and their affiliation;
  4. a brief one sentence submission description;
  5. an extended submission description (optional);
  6. all relevant citations;
  7. observations (optional);
  8. relevant discussion points (optional);
  9. type of model (signal/tissue);
  10. number of free parameters;
  11. pre-processing on the signals (if any);
  12. outlier rejection strategy (if any);

Also, we encourage submissions of models even when suited for only part of the data, as the scope of the challenge goes well beyond the pure prediction. In such case, please mention in the info.txt which points are disregarded by your submission.

The name of the zip file does not matter, however the names of the text files are critical. The predictions must be in the same format of the provided signals, with one column per signal (in the corresponding order), and one row for each measurement specified in the acquisition details of the unprovided data. Predictions of multiple acquisitions strategies (i.e. PGSE DSI-like vs PGSE multi-shell) must be submitted individually (i.e. submit each prediction independently). Although not required, we warmly welcome predictions for multiple datasets (PGSE and DDE/DODE).

Example submissions files will be made available with the data.


Each submission will be evaluated by computing the weighted averaged mean squared error (MSE) between the provided predictions and the corresponding undisclosed data. In case of N points corresponding to a single diffusion weighting (e.g. shell), these will weighted as 1 single point. Three winners will be selected independently for PGSE (shells and grids will be scored together), DDE and DODE.

How to get the data

Please see “Registration and Data Access” Page.


  1. Stejskal, E. O. & Tanner, J. E. Spin Diffusion Measurements: Spin Echoes in the Presence of a Time-Dependent Field Gradient. J. Chem. Phys. 42, (1965).
  2. Shemesh, N., Özarslan, E., Komlosh, M. E., Basser, P. J. & Cohen, Y. From single-pulsed field gradient to double-pulsed field gradient MR: Gleaning new microstructural information and developing new forms of contrast in MRI. NMR Biomed. 23, 757–780 (2010).
  3. Ianuş, A. et al. Accurate estimation of microscopic diffusion anisotropy and its time dependence in the mouse brain. Neuroimage 183, 934–949 (2018).
  4. Froeling, M., Tax, C. M. W., Vos, S. B., Luijten, P. R. & Leemans, A. ‘MASSIVE’ brain dataset: Multiple acquisitions for standardization of structural imaging validation and evaluation. Magn. Reson. Med. 77, 1797–1809 (2017).