r03ert0
commented
5 years ago Great project @ofgulban !! Maybe you could provide a link to "Documenting projects and code"? https://github.com/ohbm/hackathon2019/blob/master/Tutorial_Resources.md#documenting-projects-and-code
ofgulban
Spatial filter design based on structure tensors for mesoscopic MR images
Omer Faruk Gulban (ORCID)
Project Description
Ultra high field MRI (7 Tesla and above) allowed researchers to acquire human brain images at mesoscopic (0.1 to 0.5 mm) isotropic voxel resolutions in-vivo*. Here is an example of such image (350 micron isotropic) acquired at 9.4T scanner using a custom-design coil at Maastricht University:
There are several interesting details that appear at this resolution which are not visible in conventional in-vivo anatomical images. Such as the smaller blood vessels within gray and white matter (see the dark lines) or layers within gray matter (faintly visible in this image). Generating such images currently requires averaging across multiple repeated acquisitions. This is because the benefits of ultra high field are traded away to increase the spatial resolution at the cost of decreased signal-to-noise ratio (SNR). Consequently, repeating acquisitions to increase SNR takes a lot of time, so much so that there is no time left for acquiring functional images within the same scanning session.
In this project, I would like to test the possibility of replacing the repeated image acquisitions (to some extent) with a specific type of filtering to increase SNR. By saying specific, I mean a family of filters that make use of a tensor field derived from the images themselves. These tensors are called structure tensors.
I have selected this type of filter to satisfy a few constraints. The selected filter should be:
Here is an animation created from one of my pilot implementations on an artificially noised 7T T1w image:
I think this implementation can be improved, applied to other image types and validated further.
This project is by no means a novel implementation of such a filter (see Mirebeau et al. 2015). However, the application to ultra-high field MRI in the context of multi-echo and complex domain images might be novel. If for nothing, I think this project would be helpful for interested people to gain deeper understanding of tensor fields, their role in diffusion and insight on some of the current challenges of in-vivo mesoscopic MRI at 7 & 9.4T.
Skills required to participate
Integration
People can join by contributing to the following:
Programming: Scrutinizing code by writing test cases, optimizing for faster runtime, improving user interface (see related tutorials here).
Documenting: Improving docstrings (see tutorials), application to different cases, helping in quantification of performance against other methods.
In other ways that I couldn't think of here.
Milestones
Discuss conceptual and implementational details of the filter.
Implement the filter usable though a command-line interface.
Apply it to empirical data (e.g. 7T & 9.4T images that I will bring) and evaluate the results.
Preparation material
GitHub repository
I am planning to implement this filter as an additional feature in a small free and open source project that has a few other image processing algorithms implemented for 2D and 3D images.
Communication
Chat on gitter.