Not functional fMRI
My friend Feisal Kamil pointed me to this Popular Science article, Are Functional MRI Scans Not Functioning Right? I’ll discus the PopSci article and the journal article that is referred to.
If you’ve been following me for a while you know I have a series of posts about medical imaging ( #CHMedicalImagingSeries ). So maybe you have also seen some of my posts about fMRI ( #fMRIblob ).
Here’s the MRI post in that series if you missed it.
Medical Imaging 101 pt 3: MRI
For the new followers, fMRI is functional MRI. The equipment is the same between MRI and fMRI. So first off, the Popular Science caption for the first figure is misleading. The Siemens Magnetom Trio (http://goo.gl/wgI94) shown is a 3 Tesla clinical MRI scanner. Interestingly, Siemens software application for fMRI, syngo.MRI Neuro fMRI doesn’t list the Magnetom Trio as an applicable scanner (http://goo.gl/d3PC9). However, I’m sure it is capable of fMRI.
BOLD
When people talk about fMRI, most of the time they are referring to BOLD (Blood Oxygen Level Dependent) MRI.
More details here:
fMRI – What is love?
The Popular Science article refers to work done at Stanford by Dr. Knutson and his student Matthew Sacchet. One of the deficiencies or problems with BOLD that they focus on is spatial smoothing. Spatial smoothing is a 3D averaging method (in simple terms) to take some of the noise out of the data and make it easier to interpret. One issue that Sacchet and Knutson discuss is that new scanners have higher spatial resolution so the traditional methods of spatial smoothing (using a larger kernel) could be adding systemic bias to the data, i.e., spatial localization of striatal activity. They found using meta analysis (http://goo.gl/O5diB) that the size of the smoothing kernel could shift the foci of activation likelihood when comparing data with differing kernel sizes.
Partial volume effect
The paper states that magnetic field strength and voxel volume did not account for the bias. Since this was a meta analysis, the scans were done on different MRI scanners with different field strengths and therefore different parameters. Again, this is one of the weaknesses of meta analysis. One thing that isn’t mentioned is partial volume effect. The partial volume effect is when the voxel size is large enough to include different tissue or cell types. For example, if you are imaging the brain, you are likely to have blood vessels, neurons, etc. in a single voxel. For some imaging parameters, the blood vessels and neurons might give different imaging data, yet the voxel seen is the sum of the contents of the voxel. One might say, why not just make the voxel smaller so that it is homogeneous? The problem is that when the voxel gets too small, the signal also gets too small and you end up imaging noise.
Causation vs. correlation
The other issue with BOLD in the context of neuro imaging is that you are more or less measuring changes in blood flow. It is assumed to mean that an area of the brain that changes blood flow is activating. In a lot of these studies (not the meta analysis) subjects are asked or trained to do various tasks while the fMRI is being acquired. The idea is that enough control tasks are performed to get more at causation and not correlation.
The first image below is from the Pop Sci article. The second is an example of the Siemens syngo.MRI Neuro fMRI application. The last two are from the Neuroimage article and show the effect of kernel size. One shows a subtraction image so you can see where small and large kernel size methods agree. The other shows how the activation as a function of smoothing kernel, with the region of interest outline in gray, in the last image.
Pop Sci article: http://goo.gl/IzH9p
Stanford News article mentioned: http://goo.gl/zPT6S
Spatial smoothing systematically biases the localization of reward-related brain activity.
Sacchet MD, Knutson B.
Neuroimage. 2012 Oct 27;66C:270-277. doi: 10.1016/j.neuroimage.2012.10.056.
http://www.ncbi.nlm.nih.gov/pubmed/23110886
More #fMRIblob posts if you are interested.
When a blob is just a blob
Whale of a story
Meta-consciousness Brain
#ScienceSunday
Sorry Onkar Singh Gujral sometimes the hashtag search is wonky. When making the post today, only one of my #fMRIblob tags worked.
I’m just saying you might not find everything.
Not directly related to the fMRI algorithm issue above but still part of concerns that MRI imaging methods are not as low risk high reward {in certain special cases, Chad Haney ‘s point below about not wanting to raise unwarrented fear is well taken} as had been presented to the general public:
ProPublica has another report on the apparent problems encountered by those with serious kidney disease when exposed to the contrast agent Omniscan. Omniscan contains gadolinium-based imaging dyes that are bonded with a protective coating to keep the gadolinium inert. It is supposed to be excreted by the kidneys. In a currently ongoing trial, the plaintiff contends that , he contracted the skin ailment, known as nephrogenic systemic fibrosis, because of an injection of Omniscan. {This case originated in 2006}
http://www.propublica.org/article/burn-the-data-did-a-company-try-to-hide-risks-of-ges-mri-dye
Gaythia Weis I’ll respond in more depth tomorrow but your point can have negative consequences. The issue of NSF is well known 500+ articles and it is rare. There are two key points, you have to have renal impairment and use a particular contrast agent (OmniScan is one of them). There are so many valuable MRI scans that use zero contrast agent. I would hate for someone to avoid getting an MRI because of confusion about MRI safety.
Chad Haney I’ll look forward to your post. I do agree that MRI’s, are quite safe for most of us, giving much very valuable information. I certainly would seek to have an MRI when needed. There needs to be strong linkages between patients, personal physicians doing the referrals to specialists, the specialist themselves, and those who developed the technologies involved so that all are fully informed.
Similarly, in your example above, the researchers who understood the structures of the brain ought to have been in closer communication with the engineers who understood the signal to noise ratio of the instrumentation and the mathematicians and programers working with the algorithms.
MRI is a wonderful technological tool, but we shouldn’t let the “oh wow” factor overwhelm our need to ask pertinent questions.
NSF is so well known in radiology right now, I don’t think any radiologist is going to risk getting sued. So questions about the healthiness of a patients kidneys will be on someones check list.