I’m part of the UCL Centre for Medical Imaging and work in the Microstructure Imaging group, and in particular the Camino project. As a group. we’re interested in using advanced imaging techniques to infer details about the tissue on the sub-voxel scale in vivo, particularly in the brain.

To do this, we make a lot of use of DIffusion MR Imaging. DIffusion MRI is a form of MR imaging that is sensitive to the motion of particles on very small scales. In fact, by varying the particular pulse sequence used in the scanner we can vary the length- and time-scales we’re looking at.  The brain sits in a bath of liquid called Cerebro-spinal fluid and the molecules in this fluid are in motion. This is useful for studying microstructure because any structure gets in the way, preventing (or at least discouraging) molecules from diffusing past it. By making measurements or diffusion in different directions over different length- and time-scales, we make inferences about the shape and size of any structure present that’s getting in the way.

In free diffusion, particles spread out evenly in all directions and move in a simple, predictable way (on average!)

In this way, we can tell the difference between diffusion in free space, and diffusion inside (say) a cylinder. We can also tell which way a cylinder is pointing, as this will be the direction in which diffusion is the greatest. With enough measurements and the right analysis we can also tell how wide the cylinder is information about its neighbours.

By modelling tissue structure as combinations of different shapes or other distributions of particles, we can measure what types of structure are present in different places in the brain. What’s more, we can do this completely non-invasively in a living brain. Potentially, this means that this type of information could be used to diagnose diseases (different types of tumour have different structural properties, for example) or to plan surgery, but showing what healthy and important areas to avoid during the procedure.

Inside a cylinder diffusion is restricted across it and unrestricted along it. Measurements of diffusion therefore tell us about the cylinder.

This is not what I do. All these techniques require valdation, and new techniques need test-beds to ensure that they are as accurate as possible. In order to test the models, you need to know a ground-truth – you need to knwo what it is you’re looking at so that you can compare the results to what’s actually there. This is harder than it sounds – cutting open a brain and taking microscope images is a complex process (and has complications of its own). You could also make a physical model of the tissue whose properties you know, but brain tissue is complex stuff and this can also be challenging.

Our approach is to construct a simulation. Although they too are far from perfect, simulations enable you to construct quite complex tissue models and control the underlying physics and the assumptions you make. So hat I do is make simulations that people can use in their own studies, both in the MIG group and beyond.

The simulation models the diffusion of particles as random walks in an environment that contains structure that gets in their way. Every time a particles encounters a barrier, it bounces off (or, if the tissue is permeable, it might pass through it some of the time). Over the top of this, we simulate a diffusion MRI pulse sequence which uses the motion of the particles to construct synthetic measurements from the scanner.

Modelling diffusion with random walkers is a similar assumption to that made in most analysis of diffusion MRI data. The tissue, on the other hand, can be modelled in a number of different ways. For example, we could use cylinders or other simple shapes but the main advantage of a simulation-based approach is that we can capture some very complex structures. Taking a cue from the sort of approach you might see in 3D computer graphics programming, the Camino simulation uses meshes of triangles to model structure. This means we can model almost anything. For example, we can construct a mesh from a stack of microscope images.

This one is actually from a piece of asparagus

By constructing a diffusion substrate from a stack of microscope images, we can capture a lot of the structural complexity of biological tissues.

The red spots are diffusing particles. Where there are holes in the mesh, they can pass through. Where there is structure they are restricted. This particular sample was imaged in a 9.4T scanner before being sectioned and imaged with a microscope so we could compare the output of the simulation with the measured signals. Details of the experiment are here.

We can also vary the MR pulse sequence used to get the measurements. Optimising these can make a big difference in the sensitivity to what you’re trying to measure, much like focusing a camera and setting the exposure correctly ensures the best photograph. The simulation can be used to test new sequences before trying them out in the lab.

Of course, we don’t HAVE to use a biologically realistic sample…

DIffusion inside a hollow cow

Galleon diffusion is an under-explored problem

So by constructing the simulation we are able to support the development of new image analysis techniques and the new pulse sequences for optimal senitivity. The diffusion is freely available as part of the Camino diffusion MRI toolkit.

The images of the simulation were created using the Seer 2 Camino simulation visualiser.