There is growing interest in the neuroscience community in estimating and mapping microscopic properties of brain tissue non-invasively using MR measurements. Diffusion MRI (dMRI) has emerged as a key modality for microstructure imaging because of its unique sensitivity to cellular architecture on the scale of microns, enabling the possibility to make inferences of specific microstructural features such as neurite density, axonal diameter, axon orientations, cell sizes, and myelin fractions (Novikov et al., 2016). This has enabled scientific advances for studying normal brain development and aging, as well as better understanding a broad range of brain disorders.

Because of this, there now exist a large number of techniques to estimate white matter (WM) microstructure. These models differ in the specific tissue parameters estimated, the modeling strategies employed, and experimental design. Many of these techniques are now transitioning from the basic research domain to a wider application in biomedical studies (Alexander et al., 2017). However, in order for these to be useful biomedical tools, it is critical that they are validated to ensure that they yield accurate, precise, and reproducible results (Dyrby et al., 2018). The main challenge facing microstructure validation is the lack of anatomical ground truths. Validation is typically performed through (1) simulated data, (2) physical phantoms, (3) comparison to histology, or (4) comparisons to literature. In the past, however, validation has been limited to typically one microstructural index, with one (or few) algorithms assessed, and only one acquisition performed. Furthermore, the evaluation of an algorithm’s performance is hindered by difficulties creating realistic simulated microstructural environments, or the resources needed to perform both MRI acquisition and histological processing. Because of these difficulties, there are a number of unanswered questions in this field:

  • What model (either biophysical or signal model) better explains the underlying tissue environment?
  • What dMRI sequence enables the most accurate estimation of these parameters?
  • How does experimental design affect these measurements?
  • What is the reliability of estimating the various indices?
  • How many of these compartments can actually be estimated using dMRI?
  • How appropriate are simplified simulations and physical phantoms for microstructural validation?

To answer these questions, we present the White Matter Microstructure with Diffusion MRI Challenge. Unlike previous challenges, this will be a 2-year challenge. The project will be run in partnership with QMENTA who will provide its innovative data management platform. The first year will be dedicated to designing the challenge, building the appropriate dataset(s), and presenting it for the first time at the 2019 ISBI conference. The challenge will then take place in the second year, giving research groups and clinicians a full year to implement, and potentially revise, their submissions and algorithms.

The rationale for this is two-fold. First, this will ensure that we design the appropriate challenge to answer the questions that are relevant to the field. While the committee will steer design and evaluation, we will invite feedback from the entire microstructural field, accessing a larger pool of potential data and contacts, as well as a balanced set of opinions on how to define the reference standards and evaluate algorithm performance. Second, from our previous experience from hosting challenges (ISBI 2018, ISMRM 2017, MICCAI 2013, MICCAI 2015) participant feedback has indicated that the time between data release and submission deadline (either due to compute time, delays releasing data, time to optimize submissions) is limited, resulting in a number of submissions that may not have been properly tuned or implemented. This extended timeline for both data release and data submission gives us the opportunity to generate the perfect validation dataset, and the potential for uniquely hosting a number of “sub-challenges”, for example, we aim to host 3 sub-challenges evaluating our current ability to:

(1) predict unseen signal (signal representation; sub-challenge #1)

(2) estimate microstructural measures (signal modeling; sub-challenge #2 and sub-challenge #4)

(3) evaluate sensitivity and specificity of potential biomarkers (biomarker evaluation; sub-challenge #3).