All That Glitters Is Not BOLD For fMRI Scanning?

By Neuroskeptic | February 3, 2014 12:29 pm

A new paper warns that: All that glitters is not BOLD. This title seems designed to worry neuroscientists, because the blood oxygenation-level dependent (BOLD) phenomenon is what allows fMRI scanning to detect brain activity.

Or is it? Writing in Scientific Reports, Finnish neuroscientists Ville Renval, Cathy Nangini and Riitta Hari argue that BOLD isn’t always central to the fMRI signal.

To arrive at this claim, they examined what happens to the results of fMRI brain scans when you fiddle with certain parameters; the key experiment compared different options of the flip angle (FA).

The FA is one of the fundamental settings that must be specified for any MRI scan. However, I think it’s fair to say that most neuroscientists have not given the FA much thought, tending to stick with the default setting known as the Ernst angle.

To see if the FA matters, Renval et al showed some volunteers viewed a grey circle which, at certain points, started flashing black and white in a square pattern, like a checkerboard. It’s known that the flashy checkerboard strongly activates the brain’s visual cortex more than a static grey.

During the study, the flip angle was varied over the course of the scans – the options were 12.6, 22.5, 50 and 90 degrees and some of these changes happened within each scan, which is an unorthodox (but informative) approach. Another parameter, the echo time (TE) was also varied: 16 or 30 milliseconds.

Because the volunteers were always seeing the same two stimuli (static grey vs flashing checkerboard), we can assume that the brain was activating in the same way throughout the scan (i.e. with more activity in visual cortex to checkerboard.)

The different FA set-ups would, therefore, ideally have detected the same results. However…


Changing the FA had no effect on measured activity in the visual cortex, the first four regions pictured above. The green circles show perfect overlap between the activation maps.

But it was a different picture elsewhere in the brain. In particular, the measured activation was highly FA-dependent in the zeitgeisty ‘default mode network‘ areas, the posterior cingulate cortex (PCC) and the precuneus. This suggests that the underlying basis of the fMRI signal is not the same everywhere in the brain.

There’s more. The key finding of the paper comes in Figure 4, a behemoth of an image which I’ll display in full:


Look closely and you’ll see that the grey slice-of-brain picture is composed of hundreds of little graphs. Each has six lines, depicting the average checkerboard-evoked signal over six different conditions of FA and TE.

Renvall et al say that these data confirm “non-BOLD” behavior in many areas of the brain. The signal in many areas does not behave as BOLD signal ought to (that is, it should vary with TE, and not vary with FA.)

Why not? The stock villain of neuroimaging, head movement, did not seem to be responsible. However areas of the brain containing fluid (CSF) were found to be high in non-BOLD signal. It’s still not clear, in other words, what the ‘non-BOLD’ contribution to the activation represents… which is rather unsettling.

So is this a bombshell for fMRI? Surprisingly (given the title of their paper), Renvall et al end on a reassuring note, saying that the messiness they’ve uncovered probably cancels out in the grand scheme of things:

Our results are likely less central for traditional group-level analyses of activation locale (brain mapping) because the discrepancies tend to average out due to anatomical variability, and even localized non-BOLD changes still represent functional signal.

Still, these interesting results suggest that depending upon the specific brain areas of interest, tweaking the acquisition parameters could bring benefits for fMRI. Certainly after reading this paper neuroscientists will be less likely to be so flippant about flip angles.

You can almost see the disapproving frown on the faces of Renvall et al when they wrote:

The commonplaceness of fMRI has given the impression that this imaging method can be used in an ‘‘automatic’’ mode, without worrying about the biophysical basis of fMRI signal changes. Some recent studies do not report even the basic acquisition parameters, such as TE, TR, or flip angle (FA), as if these were well-established defaults with no appreciable influence on the data.

ResearchBlogging.orgRenvall V, Nangini C, & Hari R (2014). All that glitters is not BOLD: inconsistencies in functional MRI. Scientific reports, 4 PMID: 24472878

  • DS

    Hi Neuroskeptic

    The authors are doing the field a great service by highlighting these problems. But I think they are dismissing motion as a cause of these discrepancies without adequately justifying that dismissal in their paper. Here is the paragraph that does the dismissing:

    “Additionally, bulk head movement could induce, conceivably with a stimulus-related temporal pattern, tissue-contrast-based signal changes. While our data remain partly ambivalent to the nature of this effect, head movement parameters captured by motion correction do not support the bulk movement hypothesis, and the measured voxel timecourses are not consistent with stimulus-locked head motion,”

    In what way are the voxel timecourses not consistent with stimulus-locked motion? The authors do not say. I certainly hope that they did not simply calculate correlations between motion parameters and data. The effects of motion occur at many time scales and follow different trajectories than the motion parameters themselves. For example spin history effects evolve on time scale of many seconds (many TRs) despite the motion occurring on a much smaller time scale. The same time scale is at work for local susceptibility (time varying geometric distortions and signal dropout) motion-related effects. Clearly a correlation analysis between voxel time courses and motion parameters would not be sensitive to all motion-related error.

    I should also point out that the experiments were carried out using a 12 channel head coil and a SENSE acceleration factor of 2. It is known that images reconstructed by the usual sum-of-squares method from multi-channel receiver arrays are, upon motion correction, contaminated by a receiver field contrast that moves relative to the head. It is also known that accelerated imaging has more motion related problems due to the usual sources of motion-related noise as well as extra sources related to the applying a receive field map acquired at one point in time to an entire time series of images.

    Here’s another thing that caught my motion-sensitive attention. The paper explains that some subjects data were eliminated from consideration because:

    “Five subjects were excluded from analysis: head motion was considered too large (≥1.3 mm) in three subjects, and the time series of two subjects were contaminated by large fluctuations throughout the brain that could not be explained by cardiac and respiratory cycles.”

    Well what was the source of those “large fluctuations throughout the brain” with those two subjects? If it were obvious in two subjects could it not be present yet not obvious in the remaining subjects and thereby contaminate the findings?

    I will post most of my comments on pubpeer as well. I would really like some feedback from the authors on this extremely important topic. Kudos to the authors for highlighting the problems despite my criticisms as to the cause.

    • practiCalfMRI


  • practiCalfMRI

    There’s nothing “pure” about BOLD. It has many components – most simply, intra- and extravascular, and “small” and “large” vessel dephasing – and there are well-known (direct) dependencies on cerebral blood flow (CBF) and volume (CBV). For example, Scouten & Constable ( showed how changes in the volume fraction of CSF can alter – even nullify – BOLD changes because of concomitant CBV changes.

    The flip angle has been recognized as a generator of CBF-weighted contrast – what we usually call inflow effects – for a while. See this review,, for loads of great references and history. Exciting blood that is destined for a downstream slice is a version of perfusion weighting; a primitive arterial spin labeling (ASL) sequence.

    In a standard single-shot EPI sequence as used for fMRI, we have the following main contrast dependencies on parameters:

    Voxel size: T2*, and CSF partial volume
    Echo train length: T2*
    TE: T2*
    TR and flip angle (which are considered as a pair): Blood T1 inflow.
    Slice direction and order: Blood T1 inflow. (Probably small relative to TR and flip angle.)

    I have probably forgotten a few, too!

    The paper and this blog post are timely reminders that BOLD isn’t a universal constant. It’s a phenomenon, a term that is meant mostly to indicate T2* weighting at TE, but can also be used to indicate T2 weighting at TE in a spin echo, and other things. Does it matter? To my mind, no. It’s all vascular changes of some sort. The real task, it seems to me, is to set up the experiment so that there is robust functional signal change. Only when contrast changes are opposing/offsetting do we start to worry. One way to do that is to use a low flip angle to reduce inflow effects (20 deg at TR=2000 ms might be a good start) and then be aware of CSF volume fraction as another possible issue. Finally, low FA may be beneficial for reducing physiologic noise, too:

  • petrossa

    i’m gloating now. :) the “i told you so’s” are flowing through my brain. Good luck with rectracting your wonderful papers on how place X is involved with higher order proces Y.

    • Jona Sassenhagen

      Nothing like that will realistically happen, and even if it were, gloat would hardly be the appropriate response.

      • Neuroskeptic

        To be fair, if that did happen, Petrossa would be entitled to gloat because he has been saying it should happen for years.

        However, it won’t happen.

        • petrossa

          the retraction won’t happen that i can see too. It should but it won’t. Maybe i’ll start to compose a list of the more absurd papers.One springs to mind: The one linking rational decision making and religion. They use the same ‘centre’ ( paraphrasing from memory) Couldn’t stop laughing.

          • DS


            Which papers do you think should be retracted? Could you be more specific?

          • petrossa

            as per my previous post, i obviously didn’t compose a list of retract worthy papers since at that time i didn’t see the point.I just got annoyed and forgot about it. From my previous posts one could conclude i am of the opinion that fMRI is akin to a 48 pixel photo in resolution and as such not a tool to determine subtle higher order processes which seem to be the ‘thing’ nowadays. Not only because of its low resolution, (and the rather iffy data for statistical basis of the software) but also because it doesn’t mean much of some part is sanguineous if you don’t know what that does. Is it inhibitory/excitatory for some completely different part which is masked off because you didn’t expect anything there, just a part that’s used as a spare part for other processes etc.
            As such it is as useful as an EEG but with prettier pictures. And we all love pretty pictures, don’t we?

          • DS

            Your criticisms are too general to be of use to me so allow me to narrow the focus. There have been some interesting papers on the topic of decoding fmri data during visual stimulation of various types. What do you think of them?

          • petrossa

            I promise from now on to bookmark all papers i find exemplary of fMRI abuse. The decoding of the visual memory i find exactly proving my point. I’s a 48 pixel photograph of a predetermined object. Now show me a paper that can read random information in a way you don’t have to clean up to be recognizable … It reminds me of people ‘communicating’ with totally paralyzed people where a ‘helper’ supports the hand with pointing at letters.

  • Pingback: Magapsine (04/02/2014) |

  • DS

    Since the image contrast changes significantly with flip angle then
    couldn’t the discrepancies with respect to flip angle be explained by
    motion and contrast changes alone? If there exist regions where the
    contrast goes from low to high to low again (or high to low to high) on a
    spatial scale of residual motion (Residual after realignment due to
    motion parameter error or interpolation error) then these results would
    be expected. Am I right?

    I thank the authors again for their nice paper and I hope they will respond to my pubpeer comments. I think these results are important and the cause cries out for further investigation.

    • DS

      So far the authors are not participating in the discussion at PubPeer.

  • Wouter

    I feel that this paper strongly underlines the need for fMRI-scientists to publish the actual BOLD-signal along with their statistics. Too many papers present fMRI-results as mere T-maps and F-statistics, but are not revealing the actual (averaged) BOLD-response. I feel that this makes it impossible to attribute any real value to the statistics, since (apparently) all sorts of fluctuations are present in the data, which we actually already knew.

  • JR


    Thank you for passing this article.

    I am not at all an fMRI expert and was thus wondering whether the different sequences/parameters could lead to different auditory stimulations.

    Thank you,

    All the best

  • Pingback: All That Glitters Is Not BOLD For fMRI Scanning? | electronics()

  • Pingback: Unrelated to all that, 2/7 edition | neuroecology()

  • Pingback: That Glitters Is Not BOLD For fMRI Scanning? – Discover Magazine … | the scan book()

  • fMRI_today

    Interesting work however a bit overblown as most of what’s reported is generally not new and is understood by most fMRI methodologists. I included a flip angle study in my thesis in 1992 (p. 295, Fig 9.10) but never published it.

    Recently, we picked up on flip angle effects and demonstrated that it’s generally better to use a much smaller angle than Ernst angle as long as physiologic noise dominates.

    1. J. Gonzalez-Castillo, V. Roopchansingh, P. A. Bandettini, J.
    Bodurka, Physiological noise effects on the flip angle selection in BOLD fMRI. NeuroImage 54 (4) pp. 2764 – 2778. (2011)

    Overall, I’m all for reporting interesting variations in BOLD that occur with parameter variations, however I have an issue the delivery here: that it’s somehow mysterious, new, a problem (it’s not a problem in most cases), and of concern in how we fundamentally interpret BOLD. Not true at all and slightly disingenuous to the general MRI community.

    • Neuroskeptic

      Thanks for the comment!



No brain. No gain.

About Neuroskeptic

Neuroskeptic is a British neuroscientist who takes a skeptical look at his own field, and beyond. His blog offers a look at the latest developments in neuroscience, psychiatry and psychology through a critical lens.


See More

@Neuro_Skeptic on Twitter


Discover's Newsletter

Sign up to get the latest science news delivered weekly right to your inbox!

Collapse bottom bar