"Is that a 7 Tesla magnet in your pocket, or are you just pleased to see me?"

German neuroscientists Metzger et al report on the results of a study using the latest, ultra-high-field Magnetic Resonance Imaging to measure brain activity in response to sexually arousing stimuli.
Stronger magnetic fields mean that the scanner is able to pick up smaller differences in brain activation, with a better temporal and spatial resolution, although it's not all good news because some of the artefacts that can spoil MRI images also get worse with higher fields. 7 Tesla magnets are also so incredibly powerful that you literally have to tread carefully around them: move too fast through the field, and you can suffer dizziness, vertigo, and visual disturbances...
Anyway, Metzger et al's paper is one of the first 7 Tesla fMRI studies and it's pretty cool. They managed to achieve a spatial resolution of 1.7 x 1.7 x 3 mm , or about three times better than most studies, and a temporal resolution of 1 second, twice or three times better than usual.
They showed heterosexual male subjects a range of pictures, some of them pornographic, others "emotional" but non-sexual. They found that anticipating seeing a picture, as opposed to than actually viewing one, activated different areas of the cortex and also of the thalamus (see image above). Sexual arousal was also correlated with activity in a third thalamic area.
This fits with previous work, so it's not too surprising, but it shows the power of 7 Tesla fMRI: as you can see, the thalamus is a small structure, and conventional fMRI struggles to localize activity to particular subdivisions of it. But we know that the thalamus is a hotbed of activity, because almost all the information that goes to and from the rest of brain passes through it. Until now, fMRI researchers have tended to treat the thalamus as a no-man's land but with any luck, 7 Tesla scanners will start to change that. For those who can afford them...
Metzger CD, Eckert U, Steiner J, Sartorius A, Buchmann JE, Stadler J, Tempelmann C, Speck O, Bogerts B, Abler B, & Walter M (2010). High field FMRI reveals thalamocortical integration of segregated cognitive and emotional processing in mediodorsal and intralaminar thalamic nuclei. Frontiers in neuroanatomy, 4 PMID: 21088699
9 comments:
One second temporal resolution? That's nothing compared to EEG. Even a bad EEG recording has temporal resolution an order of magnitude better than that.
Other than that, this is pretty cool.
but I always thought higher magnetic fields e.g. 7T would resolve the trade-off between the SNR and spatial resolution most efficiently. How does the temporal resolution come into play? why should that improve?
and a follow-up on that: you can never say that the 'effect' you see compared to 1.5 or 3T is due to less noisier recordings or separate regions giving rise to the effect. can you?
@spike
They used some fairly standard tricks to decrease the time taken to acquire the data matrix (partial Fourier, GRAPPA R=2 parallel imaging) as well as a sinusoidal read gradient. Even so, the temporal resolution isn't radically different (for 16 slices) as is feasible at 3 T.
The relationship between B0 and experimental SNR is a tricky one... Many things get worse as B0 goes up, although raw SNR gets better (for many, but not all, brain regions imaged by EPI). A good rule of thumb for fMRI is to consider a linear relationship of B0 to functional SNR, i.e. 7 T slightly better than twice as good as 3 T. Depends on the task, brain regions, etc. but it's a good throwaway rule. Meaning you can trade the SNR for higher spatial res.
On the "less noisy" question, it's involved. Respiration effects get worse at B0 goes up! What you *can* say, though, is that if you have higher spatial res and better functional contrast, you can reduce the partial volume effects of inactive tissue in a voxel, and that has real benefits.
@ practiCalfMRI:
thanks for the rule of thumb. Your point is well taken about the dependency on the region as well!
My second one, was more a conceptual one though:
imagine a case for which conventional 3T doesn't show any differences between conditions A and B and 7T does. presumably with the higher spatial resolution you have in 7T, you can say whether there are overlapping populations responding to conditions A and B or not (and I am referring to the multivariate nature of the codes here!, just separate blobs I mean). This may not have been the case for 3T due to the low spatial resolution. My whole point was that 'I think' it is impossible to say whether the separation is due to the improvement in the spatial resolution or in the readouts quality i.e. SNR?
Ah, I see where you're coming from, spike. It could be possible in principle to answer your question, simply by matching all spatial and temporal parameters at 3 T and 7 T for twin experimental tests. Then, if you see the separated blobs only in the latter you know that it was because you had more functional SNR (where the noise is everything that isn't signal of interest) and not spatial resolution. This, of course, tells you nothing about why you have more fSNR; it could be a diffuse population of neurons, or focal populations that manage to overcome the 'inactive' voxel volume.
I think I also see a note of confusion: 7 T doesn't offer intrinsically higher resolution than 3 T. Indeed, one of the interesting quandaries of ever higher fields is that the gradient and RF encoding methods aren't inherently different from those at lower fields (yet), meaning the spatial resolution available isn't inherently higher, it's just that when you have more signal to start off with you are able to make images with smaller voxels (that have non-zero SNR). Make sense, sort of? Of course, once one has more SNR one could potentially devise methods to offer a real spatial resolution benefit, i.e. novel methods that wouldn't work at lower fields. I have a colleague who is doing just that at 7 T, but I'm sworn to secrecy right now. As soon as he gets his stuff accepted I'll see if he's interested in a post.
As for the TR, they used a TR of 1 second. That's certainly not impossible at 3T, there have been studies using a TR of 100 ms, but they scanned a single slice, in this study, they managed to get most of the brain.
@ practiCalfMRI:
thanks, yes, it does make some sense except that you are saying: ah! 7T has 2 advantages, let's get rid of one to make the picture clearer (matching spatial resolution in this case). To me this doesn't seem to be the best way to get the answer to the conceptual question...
@ neurosceptic:
cool, have you come across a study showing that BOLD 7T correlates significantly better with LFP or spiking activity (in say V1) than BOLD 3 or 1.5T?
this would be very nice.
spike: Not that I know of but that's not my field.
I did hear someone speculate that at 7T it might be possible to reliably image the "initial dip" in blood oxygen that precedes the main BOLD response, if that proves to be true, it could lead to more direct measures of activity, but I'm not sure if anyone's done that yet. Certainly something to watch out for though.
Hi spike, I've not read this paper but there are some good studies from Seong-Gi Kim's lab such as this one that might help:
Neuroimage. 2008 Apr 1;40(2):442-50.
Trial-by-trial relationship between neural activity, oxygen consumption, and blood flow responses.
Masamoto K, Vazquez A, Wang P, Kim SG.
Seong-Gi and others have looked at the spatial specificity of BOLD (T2 as well as T2*), CBV and CBF and have showed that CBF is as well localized as any of them can probably be. (They'd go in the order CBF = CBV > T2 BOLD > T2* BOLD.) Along those lines, many BOLD studies at 7 T use a spin echo for T2 contrast, in part to get the specificity but mostly to preserve signal in bad susceptibility regions. SE-BOLD is difficult at 3 T and below because the signal changes are so small.
If you're really into the spatial specificity stuff, Seong-Gi has also done some great laminar-specific imaging with CBF and BOLD; he can resolve input layers, for instance. (But he's using animal models at 9.4 T!) There seems to be better specificity perpendicular to the cortical surface than parallel to it, possibly because of diffuse populations of neurons across columns...
Re. the initial dip, i.e. direct monitoring of CMRO2, it's tricky, not least because the post-stim undershoot from prior trials can easily alias. Jens Frahm showed you may need to wait as much as a minute after the prior stim to be sure of avoiding aliasing! That renders the current methods of CMRO2 imaging, which are T2* based, as regular BOLD, unsuitable for routine experiments. And Seong-Gi has shown that CBF response may be just as well localized, and it's a whole lot easier!!
Post a Comment