Actually both, I don't expect to be able to buy a machine but adding travel to the cost of doing the experiments might make buying scanning time too expensive as well. In which case the cost of buying (and maintaining the machine) becomes more practical. Consider me so far out of the loop that any information will be positive.

For instance I know that MRI's in general are expensive, but are we talking millions, tens of millions or hundreds of millions of dollars? Secondly, can any MRI do fMRI or do you need a specialized machine? or specialized operators, there are local MRI's but I am not sure if there are local fMRI's. Do we need a special computer to do MVPA? or can it be done with a supercomputer. I have no doubt that we can do classification in a normal super computer, but I dimly remember there might be compatibility problems. Due no doubt to different standards.

Compatibility of the EEG elements with MRI probably means more expensive, specialized gear, but how expensive should I expect it to be. Working within a Magnetic resonance field I imagine puts very strict limits on the type of equipment that can be used. One wouldn't want induced currents shocking the subject.

Right now I am looking for ball park figures so I can evaluate what level of funding is required if I want access to the technology at different levels of Access. Later I will try to figure out what is practical, or even possible once I have looked into funding.

Graeme: “I know that MRI's in general are expensive, but are we talking millions, tens of millions or hundreds of millions of dollars?”

Allow me to risk some numbers within this reply:

According to the latest link thankfully mentioned by Dr Kamping, it seems that one can have his/her open 3T MRI scanner “Heliumom filled, fully equipped and ready for immediate use. Delivery and installation are included”, with 1000000 USD. “Construction of MRI suites could cost up to US$500,000, or more, depending on project scope” according to another site. It’s about a “secondhand” machine (2008) where it may be prudent to think about the service of a “provider of pre-owned and refurbished medical imaging systems” as pretended by http://info.blockimaging.com/bid/68146/How-Much-Does-a-Used-MRI-Machine-Cost . That provider says offering assistance “with a complete financing package. Big deals or small; fast answers, buildout funds, deposits and pre-funding are all part of the deal!” (The exclamation point is the provider’s, mine is this one!)

For the second option thankfully suggested by Dr Kamping, the provided link doesn’t give an idea about the fees but similar offer (http://www.fmrib.ox.ac.uk/ ) mentions a charge per hour long scan of £365 (580.49600 USD?) and to make use of the Centre’s central computing facilities for data analysis, this carries a yearly charge of £3000 (4 771.2 USD?). BTW, that British fmri lab was “recently awarded £8 million [12.7232 million USD?] by the MRC, EPSRC, Wolfson Foundation and University of Oxford to purchase and install new 7T and 3T leading-edge MRI systems to enable us to image brain structure and function at even higher resolution than currently possible.”

May I imagine that the reply to the above question turns around 5 million USD?!

Ok, so ball park is what I asked for, and these numbers will do.

$5-10 million for at least a 3T machine but might want to up that to a 7T.

$500,000.00 for renovations for the suite.

A helium supply.

or $600.00 USD per hour for someone else's machine.

Assuming you could charge $600.00 USD for day-light hours that is an income of

1.75 million per year if no holidays or weekends were taken, and 1.12 million a year if only week days and non-holidays were used. Depending on maintenance costs, you could expect a payback sometime in the 5th to 10th year depending on how expensive your system was. The numbers seem close enough to be reasonable.

It sounds to me as if the MR machine 3T or 7T may be standard but the fMRI is how it is set up.

Many research institutions will "rent" MR scanner time to approved outside projects. This is usually part of the requirement for NSF or NIH funded centers. A base rate for scanner time is about $350-$450 per hour at a research facility. This would be your best option for getting any data in the immediate term. fMRI is not simply putting someone in a machine a pressing the "fMRI button". The functional data is heavily massaged out of the underlying signal changes in the BOLD sensitive images. Experimental technique are critical to obtain quality data. Many early experiments in the 90's were plagued by a number of artifacts that appeared to be "real" activation.

Start on a 3T scanner. The results are 7T are nice for temporal dynamics and resolution but the artifacts from the field, susceptability, and other items can be difficult to sort through. 3T fMRI is a production mode at any place doing a number of these types of studies.

Combining EEG with fMRI is quite doable. Though I think your idea of pattern analysis with triggering may be difficult to implement. Retrospective analysis with recorded EEG and MVPA of fMRI results are quite doable. Closing that loop in a controlled fashion will depend on the type of study you are attempting.

Routinely it is possible to take data from the scanner, process it and determine some feed back you wish to change in the system. Combining that with EEG would be a challenging control problem but would be amenable in certain studies where the control loop time is not too long.

In short, the recommendation is to propose your study to a robust research center with the equipment and personnel ready to assist you in getting quality data quickly. After you've built up some expertise at your institution then I would recommend getting a good 3T or higher instrument. You can purchase a high quality 3T suitable for fMRI for 1.8-2.4 million dollars. The yearly maintenance agreement would run you about $150,000/yr. Your helium fills would be included with that. Current generation MRIs no longer have a huge thirst for liquid helium.

OK thanks for a North American perspective.MVPA is usually done in retrospect, so I assume that any "Triggering" will also be done in retrospect on the basic "functional" scan data.

The "Approved Outside Projects" classification is a bit troubling, since I come from Canada and NIH and NSF are US funding agencies. But perhaps there are centers here in Canada that offer the same type of service. Or Cross Border agreements of some sort. The main problem I see is coupling the imaging information to the EEG information so that the classifier can be triggered by the EEG signal to classify the MRI image. That might be as easily accomplished as recording a common time track and adding the EEG signal synchronized to the MRI image in the data at the super computer level.

The EEG information would have to be recorded simultaneously with the MRI data, but would not have to come together until late in the processing after the functional Massaging is mostly done.

One thing I am not sure of, is how much storage a 3T image over a 1 hour period would take, and whether I would be able to do multiple MVPA runs on the same image data without hogging the super computer.

Ideally I see a classical psychology experiment with multiple stimuli, of multiple types, presented via a computer monitor/mirror/MRI qualified speaker or headphone set, and so on, with a mismatch to negative protocol on the EEG, and linking classification to EEG signal peaks. The idea is to be able to see data flow, in the BOLD images if we then animate the classified images. I see this as being a little like you can trace the winds and storm fronts using radar to link weather to a weather map.

Your technical requirements are quite doable. Continuous EPI can generate 5 to 20 images a second. 18000 to 62000 images an hour no more than 10 or 20 gigs depending on image resolution. Your usble fraction of the data will depend on the EEG events etc. Supercomputer usage for pilot projects is reasonably accessible in the us. Also in Canada according to colleagues of mine. There are good research mri groups in every province in Canada so you should have opportunities to conduct your study.

I was hoping for a slightly finer temporal resolution, but I guess that will do. It is almost exactly the temporal resolution of the signal I need to trigger off of, which means that the signal will be easily shifted if it varies in timing too much. I would have preferred a temporal resolution at least twice the frequency of the EEG signal, is 20 images a second a standard setting or is it machine specific?

Having a provincial research group, would cut my costs significantly as I would be able to drive to most research sites here in Alberta. I imagine most sites would be in or near Edmonton, Calgary, or Lethbridge, in order to maximize usefulness of the machine.

As for temporal resolution, there is a speed-volume trade-off in fMRI. EPI images can be acquired at such rapid rates (TR of ~250ms is the fastest I've personally witnessed with any success, actually) but the field-of-view has to be very small (2-6 cm) to do this. The limiting factors are the tolerance/capabilities of gradient coils and T2 (spin-spin interactions) related effects that can cause susceptibility artifacts at such rapid acquisition. It is unlikely that you will be able to get large cortical areas with high speed EPI.

The utility of long (1 hour) simultaneous EEG-fMRI is also somewhat hampered by the duration of the BOLD signal and scanner stability. Rapid ("real-time") fMRI can estimate susceptibility changes consistent with BOLD activity, but requires a great deal of processing to differentiate the signal from noise characteristics (again, the smaller the FOV, the better). It appears that you would like to create an index of BOLD dynamics with different EEG paradigms, and that sounds very exciting. As an FYI: You will definitely need to be careful with the magnet SNR stability for the scans as this may vary as the gradients heat up over the duration of the session.

While the technical hurdles are somewhat large, EEG-fMRI offers an exciting opportunities for research that I hope will continue to receive more attention in the field.

U of Alberta has an excellent MRI group that should be a good partner in your study. As for imaging rate - the previous poster Keith - has an excellent point. The rate is really dependent on all the factors your require out of the image. It sounds like you are familar with concepts such as bandwidth. The bandwidth of the imaging experiment is dependent on the SNR you require out of the resultant. This is another factor that would be sorted out once you've started working with a local MRI group. Good luck with your study.

Indeed, The Peter S. Allen MR Research Centre offers interesting facilities http://www.invivonmr.ualberta.ca/Resources/facilities.php . Also sustaining the above valuable replies , this document explores some challenges and limitations of fMRI http://culhamlab.ssc.uwo.ca/JodyCulham/Courses/Neurosci500/FunctionalMRI_Constable.pdf .

OK, the lower limit of 40 milliseconds per epi scan is troubling, but I can see where the physics almost requires it. Obviously that is going to require a fairly tight FOV creating a situation similar to the FOV limitations that affect High powered micro-scopes. By the same token, higher resolution voxels seem to reduce SNR so there is a limit to resolution depending on the field strength.

The Peter S. Allen MR research Center has some interesting facilities as Abdessamad suggested, and at $525 per hour with a 30 minute minimum with no difference in cost between the 1.5T, 3T, or 4.5T machines it falls well within the ball park figures found earlier. $450/hr USD to $600 USD equivalents european. However my project might not pass the filter, if only because I am not an academic, or clinical user.

I have contacted the center anyway, and will report back any response from them.

The particular EEG traces I want to trigger off of might work better at 30 milliseconds temporal resolution, there is likely to be some registration error at 40. But considering that the BOLD field susceptabilities are so drastic, it makes some sense not to push, although using a spiral design might offer some speed reductions, it also results in a much more difficult mapping regimen.

We have been using MRI to perform gene expression profiling in living mice. It has sensitivity no less than RT-PCR and it is in whole brain. although we have detected a cohort of progenitor cells (0.03 mm3) or in vivo hybridization in living mouse, the sensitivity compared to in situ hybridication is dependent on the field strength of MR. Hopefully, the sensitivity for in vivo hybridization can improve when MR hardware and software is improved.

OK, the Peter S. Allen MR Center got back to me, and they have recommended a psychologist that pioneered EEG with MRI at the center, I have contacted him and am waiting for a reply. There are forms to fill out, and Ethical Approval to get.

Now I wish you good luck!