Instructor Notes
Lesson motivation and learning objectives
This lesson is designed to introduce learners to the analysis of diffusion Magnetic Resonance Imaging (dMRI) data using primarily Python. Although it is designed for learners who have no prior experience with diffusion MRI, a basic understanding of MRI and basic command-line and Python skills are required.
Diffusion MRI is a feature-rich modality. Processing dMRI data involves numerous aspects, and can take a non-negligible amount of time. Being able to critically assess the obtained features requires practice, usually with heterogeneous data.
Upon completion of the lesson, learners should be able to know the purpose and value of acquiring and analyzing diffusion data, as well as being able to process and analyze their own diffusion data. This lesson does not cover the details about the different dMRI data acquisition sequences. Similarly, this lesson does not cover quality assurance or checking aspects of the processing, neither does it focus on the importance of visualization for that purpose.
Lesson design
Introduction to Diffusion MRI data
- If your workshop includes the Introduction to MRI and BIDS lesson, learners will have the necessary knowledge to better understand how diffusion images are generated and different from other MRI data acquisitions. Similarly, they will be aware of the reasons that prompted the neuroimaging community to agree on a common storage convention for the data.
- Be sure that learners understand the physical phenomenon diffusion MRI is sensitive to, and how diffusion is able to capture it.
- Explain that there are other software and tools, such as 3D Slicer, DSI Studio, ExploreDTI, MRtrix, or TrackVis, to analyze or visualize diffusion MRI data.
Preprocessing dMRI data
- Pre-processing in dMRI depends on the available data (acquisition) and the quality of the data, so learners should be encouraged to look at their data to identify artefacts.
- Similarly, learners should embrace the importance of exploring the output at each step in order to ensure that the quality of the data has improved after the pre-processing step.
- Some pre-processing steps in this lesson make use of tools (such as FSL or ANTs) that are used as command-line tools, so learners should be encouraged to check their documentation, and adjust the arguments as necessary.
Diffusion Tensor Imaging (DTI)
- Learners should be able to understand the use, relevance and limitations of the DTI model, both from the a clinical point of view, and a research setting.
- Be sure to emphasize that even if the DTI model is unable to solve fiber crossings, the concepts and derived scalar maps are still relevant in practice.
- Pay special attention to the visualization of the DTI model results in order to ensure that the anatomical and the diffusion MR images are correctly registered.
- Also, stress the fact that DTI is a model applied to diffusion MRI, and that DTI must not be mistaken with the modality itself (with the acronyms DWI and DTI mistakenly being used interchangeably, partially due to the extensive use of the DTI model in clinical practice).
Tractography
- Make sure to explain the difference between the actual biological white matter fibers and streamlines in tractograms, and why tractography is not quantitative in terms of connectivity.
- Emphasize the limitations of tractography in downstream tasks, and the need to post-process tractograms to reliably account fo the white matter anatomy.
- It may be necessary to let the learners know about the potential inconsistencies in tractography file formats. These may be due to different conventions when choosing the origin or center across tools. At times it is necessary to visualize the tractograms using different software.
Concluding remarks
- Try to provide the learners with a clear idea of where diffusion MRI sits among the rest of MRI or neuroimaging modalities.
- Make sure that all learners are able to follow the lessons and acomodate the pace to the audience. Diffusion MRI being such a feature-rich imaging modality, it is important that the explained concepts are clear to the learners before continuing to the next aspect. Be prepared to be flexible and skip some aspects if necessary.
- Although the exercises are concise, be generous in the time allocated so that if part of the audience requires additional explanations, helpers can assist them during that time.
- When possible, provide remarks about avenues that diffusion MRI researchers may be exploring to gain further insight from dMRI data and overcome the limitations of current methods.
Technical tips and tricks
- Be clear about the purpose and convenience of using JupyterLab to teach the lesson. Allow some time at the beginning of the day to provide an overview of how using the learned tools would be translated to a more formal dMRI data analysis setting once a research aspect or analysis has been consolidated.
- dMRI datasets can have a relative large size (given that they
consist of several 3D volumes). Also, dMRI analysis is a computationally
heavy process, and thus, memory can become a limitation. Note that Binder services are limited to between
1 and 2 GB of RAM according
to their official docs.
JupyterLabdoes not shut down the associated kernel session when disposing (closing) a tab (usually containing a notebook) by default. Thus, the associated memory is not freed, and counts towards theBinderlimit, until the associated kernel is shut down. - Provide a broad overview of the software tools and packages used throughout the lesson, and the convenience of each of them.
- Be sure to link the tools with their corresponding documentation pages. Being at ease reading the documentation and understanding what parameters mean will be essential when learners will analyze their own data back at their institution or research works.
- If time permits, use mistakes as teaching moments: the most vital skill you can impart is how to debug and recover from unexpected errors.
Common problems
None reported.