In this interview with, Prof Richard Bowtell, Head of the Sir Peter Mansfield Imaging Centre at the University of Nottingham, he talks about opportunities and challenges of Quantitative Susceptibility Mapping (QSM), and its prospect towards clinical application.
Richard Bowtell, Prof
University of Nottingham
QSM images look absolutely stunning. However, apart from aesthetics, what makes this contrast interesting to you?
The interest is linked to the sources of contrast and the fact that we are so sensitive to iron in tissue, particularly in brain tissue, but also to myelin.
Iron is associated with many biological processes but it seems that iron dysregulations are also associated with several neurodegenerative diseases. This is still an area of very active research and we do not completely understand how much of the iron dysregulation effect is causative or a consequence of disease. QSM is exquisitely sensitive to iron and from comparison with post-mortem data it seems that the quantitation of iron seems to be pretty good. This opens up prospects for MRI in terms of diagnosis and looking at therapy.
The effect of myelin is maybe more interesting to a physicist than to the clinician at the moment. We still do not entirely understand the origin of the diamagnetic effect of myelin, but clearly there is a richness in the data. For example, we know that there is an orientation dependence and it seems that there is the possibility to use susceptibility-sensitive measurements to look at changes in myelination and diseases which causes demyelination.
Apart from that, QSM is increasingly being applied outside of the brain where it opens up possibilities for looking at bleeds or iron content of the liver for example.
You have mentioned iron and the importance for clinicians – in what form is the iron which we see in QSM in the body?
In general, we are only looking at the storage forms of iron. If we could see free iron that would be great! The main storage form of iron in the body is ferritin and it seems that this is also the dominant contribution to what we see in brain images. Of course, you are sensitive to other forms of iron for example in haemoglobin and interestingly also in the form hemosiderin, which is a decomposition product of blood after a bleed. That has already opened up the possibility of looking with accuracy at bleeds. Further, there is a possibility that there could be magnetite present. However, it seems that we are probably not sensitive enough to see that in MRI – we need other techniques.
Does the orientation of the vessels and their substructures play a significant role or is that small relative to the effect of iron?
The vessels and their orientation have a significant effect. If you were to characterize everything properly, then you would need to have a resolution which picks out the vessels in a way that you can separate the field perturbation from inside and outside the vessel. Obviously then there would still be the challenge that the susceptibility effects of deoxygenated haemoglobin in blood is very large and the signal decay times are relatively short, which makes it difficult to properly characterize.
If you do not pick up the signal from the vessels, in the case of a random capillary network their susceptibility effect is still going to be reflected in a correct way. However, when the vessels are oriented coherently then what you measure will depend on that orientation and you will not estimate the susceptibility effect properly. That is still an interesting area of research.
Quantitative susceptibility maps are calculated from MR phase images. What does it take to get a good phase image in the first place?
You need to be able to measure a complex image. The challenge hereby is to combine the signals from multiple receiver coils while maintaining the phase information in a sensitive manner. Then there is the challenge that the phase that we measure is not just affected by magnetic field effects but by effects which may not scale with echo-time. We can eliminate those by looking in multiple echo measurements and picking out the contribution that scales linearly with echo time. Assuming you can combine the signals from different coils and get rid of those non-linear phase effects, then you can produce beautiful images. These tell you something about anatomy but also show you the effects of external interfaces and field variations coming from sources outside of the region of interest. In the brain these are mainly caused by the sinuses, but also by respiration which give you time varying signatures in the phase. To get rid of the static effects, you can use filters that remove phase variation arising from external sources. To eliminate effects of things such as respiration, however, you need to have some kind of field monitoring or another way of estimating temporal field variation, then you can either take it out in postprocessing or take it out as source by driving your shims or gradients to overcome those variations.
But there the job is not done yet. Turning phase images into a QSM image is a difficult process – there are many algorithms and some work better than others. How do you see the future of this? Will there be some kind of standardization? And could artificial intelligence play a significant role here?
We have been thinking about a kind of standardization for a few years. However, it has probably not happened yet. The problem with QSM reconstruction is that if we take the phase information as coming exclusively from magnetization of the voxel, then there is a missing amount of information in k-space. What we do then is trying to estimate that information and using approaches where we can add in some priors. Furthermore, there is still the issue of how sensitive we are to the local distribution of whatever is causing the field perturbation. Maybe we can learn that from looking at other contrasts and feeding that information in. It feels like QSM is an area where artificial intelligence could help without having to stretch things too much, and the work that has come out with deepQSM and other approaches look really promising.
QSM benefits from a higher field strength. You also mentioned the potential applications in the clinic. Do you think QSM could be a driver to bring ultrahigh field systems to the clinics?
Yes, I think QSM alongside the increased sensitivity of magnitude images to R2* decay is a strong driver. It is a clear win for high field because the benefit is there, not just for physicists who like making high-resolution images, but hopefully for the clinical application of these images too. Obviously, this is linked to the importance of iron. That evolving story could make more of a difference if the role of iron accumulation in disease is better understood.
Today, QSM is not a routinely done in the clinics. What do you think is still needed to bring value to the patient?
Something that would make a big difference is QSM at the scanner, which is unavailable for many as a product. However, being able to acquire an image and to see the results on the scanner would make a difference in terms of the way things are reported, perceived and used. And of course, the standardization. Being able to compare QSM data from multiple fields and vendors in a simple way and knowing you are comparing apples with apples and not with pears – that will make a difference for sure!