Introduction

The Diffusion-Simulated Connectivity (DiSCo) Challenge is a connectivity estimation challenge presented at CDMRI, a workshop organized at MICCAI 2021. The goal of the challenge is to estimate the connectivity of a synthetic phantom composed of 16 regions of interest (ROI) connected by interlacing fiber bundles. The participating teams must provide their own connectivity matrix and the team with the highest correlation coefficient with the ground-truth matrix will be declared winner.

Motivations

The estimation of brain structural connectivity from diffusion-weighted magnetic resonance (DW-MR) images is still an open subject and depends on many steps. Roughly, connectivity estimation from a set of DW-MR images acquisitions goes as follows:

  • Image denoising and resampling
  • Signal processing and modelling
  • Tractography
  • Tractogram filtering and post-processing
  • Connectivity estimation

Because each step has its own set of methods and each methods has its own set of parameters, it is primordial to evaluate the performance of the pipelines currently used.

Data

Left: ROIs. Middle: Strands. Right: Connectivity matrix.

To do so, the organizing team has released three versions (training, validation and testing) of the dataset, each with the same ROIs and overall shape but different connectivity matrices. Each dataset was generated by deforming large tubular fibers to form arcs between ROIs, with energy parameters controlling the length, curvature and overlap with others. Each strand was then separated into smaller virtual axons of varying diameter.

The virtual axons were then used to compute intra and extra axonal water compartments, which were then used to generate a virtual DW-MR signal.

The training dataset provides the ground-truth connectivity matrix, a map of the ROIs, the noiseless DWI, a noisy version of the DWI (SNR~30), the “white-matter” mask, the actual virtual strands and more. The validation datasets provides the noisy (SNR~30) DWI, a map of the ROIs and the ground-truth connectivity matrix, and the testing dataset provides the noisy DWI and a map of the ROIs.

A script to compute the correlation coefficient between two matrices has also been provided.

Timeline

Data was released at the beginning of June 2021. Registration closes on September 1st 2021 and the submission deadline is on September 13th, 2021. The results will be presented at CDMRI’21 on October 1st 2021.

Links:

  • Challenge website: http://hardi.epfl.ch/static/events/2021_challenge/index.html
  • CDMRI page for the challenge: http://cmic.cs.ucl.ac.uk/cdmri/challenge.html