Brain-Imaging-Center / TMS-fMRI

Technical developments for simultaneous TMS-fMRI
1 stars 1 forks source link

Do we have a leakage current problem? #16

Open BenInglis opened 8 years ago

BenInglis commented 8 years ago

In Nov 2015, Daniel noticed a section in the MagVenture user guide describing an optional leakage current reduction device for use with the MRi-B91 coil. In subsequent emails from Yordan Todorov at MagVenture (pasted below) we found out that we should expect about 10 uA leakage current with our existing setup and that it could be reduced below 1 uA with the optional device. So the first question is whether we think we have a leakage current problem and should spend the money ($10,500) on the optional reduction device. Recent phantom results from Derek and Justin are suggestive of a leakage current problem.

---------- Forwarded message ---------- From: Yordan Todorov yt@magventure.com Date: Fri, Nov 6, 2015 at 11:06 AM Subject: RE: leakage current To: Daniel Sheltraw sheltraw@berkeley.edu

Hi Daniel,

Your stimulator has the Standard leakage current reduction circuit installed as part of the MRI compatibility upgrade kit.

This is sufficient when one performs interleaved TMS/fMRI and the TMS stimulus is placed between two EPI sampling. The standard reduction circuit reduces the leakage current down to 10uA.

However if you want to run interleave fMRI designs where the TMS stimulus is to be applied between the acquisition of two consecutive slices then we offer a dynamic leakage current kit that is more advanced and reduces the leakage current even further to below 1uA. This upgrade kit is offered under part number 9016E1051 (see page 12 in the user guide) and could potentially be installed in addition to everything else. I have developed and attached a quote MYT485 outlining this module for you reference.

Please review let me know if you want to proceed with purchase.

Best regards,

Yordan

DobyRahnev commented 8 years ago

Ben, can you give a little more detail about what makes it seem that there is a leakage current problem and why we hadn't seen it in previous phantom tests?

BenInglis commented 8 years ago

Hi Doby, I'm just waiting for Justin to let me know his github username so he can join this discussion and provide details. (I'll also ask Derek to set up an account.) But the short answer is that both Justin and Derek are seeing statistically significant differences between slices that have a TMS pulse versus those that don't, and the differences also appear for some slices prior to the TMS'd slice. We had previously suspected mechanical effects (waves) because of a water phantom, but both the timing of the effects and the same results in the gel FBIRN phantom suggest that we might be looking at leakage current, i.e. a build-up of the current prior to the TMS pulse that can cause regions of signal decrease or increase under the TMS coil depending on the shim. The effects are small but they are rather predictable.

derekevannee commented 8 years ago

Hi Doby, Here's what I observed: I have been applying TMS during what I believe to be the first crusher. The effect on signal change was not dramatic - with mean signal of 300-400, the mean increase in signal due to TMS was less than 1. However, t-tests revealed the effect of TMS to be highly significant. Depending on when exactly I pulsed, it could affect the slice that was TMSed, as well as a slice or two following. By playing with the timing, I was able to find spots where there was not a mean increase in signal. It remains possible that the variance is increased, but I haven't had a chance to look further.

In vivo, I find that the effect is non-uniform. In data I acquired on myself, I see a significant signal decrease in the vicinity of the coil, and an increase elsewhere.

I am about to travel for a week and I won't be able to tend to work matters during that time, but I will provide some more detailed documentation when I return.

DobyRahnev commented 8 years ago

Thanks for the update. Obviously you guys should decide what to do at UCB. It seems to me that it's worth the extra money, and I'll definitely think about buying it at Georgia Tech. Can we make any guesses as to what the alternative is? One would need to introduce a gap but how long of a gap is enough? 100ms, 200ms, 500ms? On the other hand, there's a concern whether reducing current leakage from 10uA to 1uA is good enough if we want to keep targeting the crushers (sound like it but the signal contamination may be nonlinear).

derekevannee commented 8 years ago

I, for one, would like to develop some sequences with a gap so we can test them out and compare. I’m wondering whether this dancing around the artifacts business is more trouble than it’s worth. We need 1) microsecond precision, 2) to hunt for obvious artifacts, 3) thoroughly test for non-obvious artifacts, 4) potentially additional hardware to reduce the leakage. My guess is there would be some researchers interested in a little less baggage.

From: Dobromir Rahnev [mailto:notifications@github.com] Sent: Tuesday, May 17, 2016 4:09 PM To: Brain-Imaging-Center/TMS-fMRI Cc: derekevannee; Comment Subject: Re: [Brain-Imaging-Center/TMS-fMRI] Do we have a leakage current problem? (#16)

Thanks for the update. Obviously you guys should decide what to do at UCB. It seems to me that it's worth the extra money, and I'll definitely think about buying it at Georgia Tech. Can we make any guesses as to what the alternative is? One would need to introduce a gap but how long of a gap is enough? 100ms, 200ms, 500ms? On the other hand, there's a concern whether reducing current leakage from 10uA to 1uA is good enough if we want to keep targeting the crushers (sound like it but the signal contamination may be nonlinear).

— You are receiving this because you commented. Reply to this email directly or view it on GitHub https://github.com/Brain-Imaging-Center/TMS-fMRI/issues/16#issuecomment-219880898 https://github.com/notifications/beacon/AShG2C8P1BcJaZGePl1eTg-BiKoKb3KJks5qCkqjgaJpZM4If0dO.gif

BenInglis commented 8 years ago

If the problem we're seeing is due to leakage currents then the only solution is to reduce the amount of current. There's no timing solution, it's not a timing problem. If a current leaks into the coil while the capacitor is charged then it induces a new magnetic field in the magnet and this perturbs the signal. If you never pulsed the TMS but had the leakage current you'd see the problem. Of course, in reality the amount of leakage current changes depending on use of the TMS pulses, meaning that it is entirely possible the amount of leakage current is better or worse immediately after a TMS pulse. (The leakage rate would be important.) But unless we could determine an entirely predictable, stable behavior in the leakage current, there wouldn't even be a good way to establish a "better" timing versus worse timing. It would change for sure if the TMS pulse pattern/frequency were changed, and it would likely vary with TMS coil use because the resistance of the TMS coil changes with temperature.

Final prosaic note: if it weren't a problem they wouldn't offer a $10,500 option to reduce it!

BenInglis commented 8 years ago

"Can we make any guesses as to what the alternative is?" I don't see an alternative, it's a design feature of the MagPro driving box rather than anything between the TMS and the scanner.

"One would need to introduce a gap but how long of a gap is enough? 100ms, 200ms, 500ms?" No delay will be enough, except that if one had a stable leakage current and never pulsed the TMS then it would be possible to re-shim and compensate for it as well as possible. But once the TMS is pulsed and the leakage current changes afterwards then it's a dynamic situation.

"On the other hand, there's a concern whether reducing current leakage from 10uA to 1uA is good enough if we want to keep targeting the crushers (sound like it but the signal contamination may be nonlinear)." This problem is orthogonal to where we are putting the TMS pulses. It exists all times the TMS unit is charged and ready to pulse. But we can make a prediction that if the signal changes of 1-in-300 occur now we should expect 0.1-in-300 with the new hardware. Any residual statistical significance might have to be dealt with using regressors, since the timing of the leakage current effects should differ from any hemodynamic effects.

BenInglis commented 8 years ago

J Magn Reson Imaging. 2009 May;29(5):1211-7. doi: 10.1002/jmri.21749. Image artifacts in concurrent transcranial magnetic stimulation (TMS) and fMRI caused by leakage currents: modeling and compensation. Weiskopf N1, Josephs O, Ruff CC, Blankenburg F, Featherstone E, Thomas A, Bestmann S, Driver J, Deichmann R.

PURPOSE: To characterize and eliminate a new type of image artifact in concurrent transcranial magnetic stimulation and functional MRI (TMS-fMRI) caused by small leakage currents originating from the high-voltage capacitors in the TMS stimulator system.

MATERIALS AND METHODS: The artifacts in echo-planar images (EPI) caused by leakage currents were characterized and quantified in numerical simulations and phantom studies with different phantom-coil geometries. A relay-diode combination was devised and inserted in the TMS circuit that shorts the leakage current. Its effectiveness for artifact reduction was assessed in a phantom scan resembling a realistic TMS-fMRI experiment.

RESULTS: The leakage-current-induced signal changes exhibited a multipolar spatial pattern and the maxima exceeded 1% at realistic coil-cortex distances. The relay-diode combination effectively reduced the artifact to a negligible level.

CONCLUSION: The leakage-current artifacts potentially obscure effects of interest or lead to false-positives. Since the artifact depends on the experimental setup and design (eg, amplitude of the leakage current, coil orientation, paradigm, EPI parameters), we recommend its assessment for each experiment. The relay-diode combination can eliminate the artifacts if necessary.

DobyRahnev commented 8 years ago

Ben, these are rather persuasive arguments. I always thought that MagVenture's "standard" package included current leakage control (the paper you mentioned used MagStim's system) but wasn't aware that they have 2 levels of leakage control. I feel that they should have been more explicit about this from the very beginning. On my end, I'll contact Yordan about getting the optional reduction device. 10K is not that much given how much everything else costs and how much it will cost to do fMRI tests to understand better the current leakage.

derekevannee commented 8 years ago

Does leakage occur only after delivering a TMS pulse, or can changing the pulse intensity also induce leakage?

From: BenInglis [mailto:notifications@github.com] Sent: Wednesday, May 18, 2016 6:40 AM To: Brain-Imaging-Center/TMS-fMRI Cc: derekevannee; Comment Subject: Re: [Brain-Imaging-Center/TMS-fMRI] Do we have a leakage current problem? (#16)

"Can we make any guesses as to what the alternative is?" I don't see an alternative, it's a design feature of the MagPro driving box rather than anything between the TMS and the scanner.

"One would need to introduce a gap but how long of a gap is enough? 100ms, 200ms, 500ms?" No delay will be enough, except that if one had a stable leakage current and never pulsed the TMS then it would be possible to re-shim and compensate for it as well as possible. But once the TMS is pulsed and the leakage current changes afterwards then it's a dynamic situation.

"On the other hand, there's a concern whether reducing current leakage from 10uA to 1uA is good enough if we want to keep targeting the crushers (sound like it but the signal contamination may be nonlinear)." This problem is orthogonal to where we are putting the TMS pulses. It exists all times the TMS unit is charged and ready to pulse. But we can make a prediction that if the signal changes of 1-in-300 occur now we should expect 0.1-in-300 with the new hardware. Any residual statistical significance might have to be dealt with using regressors, since the timing of the leakage current effects should differ from any hemodynamic effects.

— You are receiving this because you commented. Reply to this email directly or view it on GitHub https://github.com/Brain-Imaging-Center/TMS-fMRI/issues/16#issuecomment-220028842 https://github.com/notifications/beacon/AShG2EAdseBPRJn4lLK0JTMHHI30XeDVks5qCxa_gaJpZM4If0dO.gif

BenInglis commented 8 years ago

Derek, I suspect that the leakage current is likely to vary with the stimulus intensity setting. The presence of any sort of leakage current will almost always depend on whether the control box is charged - green light on - so that the components inside the magnet room (the manual discharge button) would be operational. At that point the capacitor is charged up and all but the pulsing switch, whatever that is, is ready to go. Some current must be leaking across that switch so that the TMS coil isn't fully isolated from the capacitor. It looks to me like the only way, presently, to eliminate all leakage current is to have the capacitor fully discharged, either because it was just used to deliver a TMS pulse or because the unit is "off" where "off" means no charge in the capacitor. What's needed, then, is better isolation of the switch between the capacitor and the TMS coil. This is presumably what we will be getting in the new box from MagPro (the purchase of which has been approved, btw).

derekevannee commented 8 years ago

I was able to look at this in a bit more detail today. I can confirm that upon delivering a TMS pulse, there is increased signal in slices that follow. The effect is most pronounced in the slice collected immediately after (not during) the TMS pulse and drops off over time. By 3 slices after the TMS pulse, I no longer saw the effect (TR = 2000, 30 slices, ~66.67ms per slice). This suggests that the effect may take on the order of 200ms to dissipate.

I delivered pulses during the first slice of a TR on some runs and the third slice on others. I’d like to look at it more systematically, but the effect seemed more pronounced for the third slice. It’s not presently clear to me why that would be the case.

The artifact is small. It seems to be on the order of 0.1% signal in the worst cases that I saw.

Let me know if you’d like me to upload some slides of these findings somewhere.

Best,

Derek

From: BenInglis [mailto:notifications@github.com] Sent: Wednesday, May 18, 2016 6:40 AM To: Brain-Imaging-Center/TMS-fMRI Cc: derekevannee; Comment Subject: Re: [Brain-Imaging-Center/TMS-fMRI] Do we have a leakage current problem? (#16)

"Can we make any guesses as to what the alternative is?" I don't see an alternative, it's a design feature of the MagPro driving box rather than anything between the TMS and the scanner.

"One would need to introduce a gap but how long of a gap is enough? 100ms, 200ms, 500ms?" No delay will be enough, except that if one had a stable leakage current and never pulsed the TMS then it would be possible to re-shim and compensate for it as well as possible. But once the TMS is pulsed and the leakage current changes afterwards then it's a dynamic situation.

"On the other hand, there's a concern whether reducing current leakage from 10uA to 1uA is good enough if we want to keep targeting the crushers (sound like it but the signal contamination may be nonlinear)." This problem is orthogonal to where we are putting the TMS pulses. It exists all times the TMS unit is charged and ready to pulse. But we can make a prediction that if the signal changes of 1-in-300 occur now we should expect 0.1-in-300 with the new hardware. Any residual statistical significance might have to be dealt with using regressors, since the timing of the leakage current effects should differ from any hemodynamic effects.

— You are receiving this because you commented. Reply to this email directly or view it on GitHub https://github.com/Brain-Imaging-Center/TMS-fMRI/issues/16#issuecomment-220028842 https://github.com/notifications/beacon/AShG2EAdseBPRJn4lLK0JTMHHI30XeDVks5qCxa_gaJpZM4If0dO.gif

derekevannee commented 7 years ago

Looking for an update here - did the dynamic leakage current upgrade solve the problem or does it persist? If it's persisting, what are the current hypotheses regarding the cause?

BenInglis commented 7 years ago

Hi Derek, We are now investigating other "hangover" mechanisms, including vibration, and trying to work out ways to isolate the different effects since it is not possible to separate them once one is doing MRI. All the effects act simultaneously! Unless and until we understand the relative magnitudes of the effects we can't easily determine how much the leakage current upgrade helped, except as you and Justin already assessed in your before/after upgrade measurements. (IIRC the statistical errors reduced by about an order of magnitude.)

At present we suspect vibration may now be the limiting factor, assuming the leakage current has been reduced to a secondary effect. And we are looking at the contribution of the asymmetry in the bipolar pulse as a major contributor to vibration. Perhaps a more symmetric pulse would have a reduced vibration profile. We're in daily contact with Yordan and others to try to understand what we've got, and to try to devise some tests. Sadly, as things stand even if we are able to rank the problems and get a better sense of the vibration contribution and other mechanisms to the TMS "hangovers," we will have limited scope to fix things in the current setup. MagPro doesn't have the facility to tweak the bipolar pulse control, for example. Once we know more we can try to persuade MagPro to work with us on solutions.

For the time being I think that you should continue to either include a dead period per TR and TMS during that, or target a slice of no interest, to be followed by a second slice of no interest, and ensure there is a delay of around 100 ms from the TMS pulse before the first slice of interest.

derekevannee commented 7 years ago

Hi Ben, Thanks for the update. I'll focus on gaps for the time being then.