I'm skeptical of anything claiming there are only 2 mechanisms driving neurovascular coupling. This seems like solid work that better characterizes 2 of the mechanisms. For fMRI I feel we need to assume we measure are an average of many factors & try not to assume a specific factor drives results.

tedana, multi-echo fMRI denoising software, newsletter includes tedana v26.0.3 updates and details of multi-echo fMRI ed course at OHBM 2026. groups.google.com/g/tedana-new...

This makes me think of "Franchise" by Isaac Asimov, which felt like an extreme thought experiment when I first read it and now seems a bit closer to real events. www.astro.sunysb.edu/fwalter/HON3...

While tedana is designed to remove noise that "should" increase pBOLD, pBOLD can test if tedana is performing as designed. Also multi-echo denoising is an active research area & pBOLD can be used to quantify the relative stability & efficacy of multi-echo denoising approaches. 8/8

Subplots a & b from figure 7 of https://www.biorxiv.org/content/10.64898/2026.03.19.712948v1 Bar graphs of prediction accuracy for session 1 and 2. NORDIC doesn't alter prediction accuracy. GSR reduces prediction accuracy and tedana increases prediction accuracy.

We also test how pBOLD realted to a measure of potential interest by using Connectome Predictive Model of IQ for each denoising pipeline. Prediction accuracy increases with tedana, but decreases with GS, implying GS is removing neurally-relevant BOLD info. 7/8

While Global Signal reduced pBOLD, a Q is if this is because GS removes phys noise or signal of interest. We show that GS is BOLD-weighted, but is not well modeled by respiratory and cardiac regressors, implying the BOLD weighting is not just phys. 6/8

Figure 5 from https://www.biorxiv.org/content/10.64898/2026.03.19.712948v1 Left side is a scatter plot of runs of TSNR vs pBOLD. The top left quadrant is low pBOLD and high TSNR and the bottom right quadrant is high pBOLD and low TSNR. The right side of the figure show seed-based correlations from 6 example runs from those two quadrants & showing the effects mentioned in the main text of this post.

pBOLD complements existing quality metrics. For example, we can identify potentially problematic runs in a dataset that have low pBOLD yet high TSNR that show scanning artifacts and other runs with low TNSR and high pBOLD that may have BOLD-weighted physiological artifacts. 5/8

Figure 4 of https://www.biorxiv.org/content/10.64898/2026.03.19.712948v1 Bar plots of TSNR and pBOLD for Basic, GSR, and tedana pipeline with and without NORDIC.

A key use of this metric is to compare denoising pipelines that can be applied to each echo separately. We compare pipelines using basic (motion), GSR (basic+global signal regression), tedana, & NORDIC. NORDIC improves TSNR but not pBOLD. pBOLD decreases with GSR & increases with tedana. 4/8

pBOLD is a measure of how closely the data are modeled by the BOLD dominated vs non-BOLD slopes. We show this method captures the non-BOLD nature of cardiac-gated fMRI and more temporally consistent volume spacing results in greater pBOLD. 3/8

Figure 1 from https://www.biorxiv.org/content/10.64898/2026.03.19.712948v1 Four subplots with scatterplots of covariance between pairs of echoes for the same ROIs. The left two plots show a slope of 1 for simulated data and very close to 1 for real data. The right two plots show a slope that is the ratio of the differences in echo times for simulated and real BOLD dominated data.

The underlying theory is based on the observation that covariance between ROIs be tween one pair of echoes vs another pair of echoes will have a slope of 1 for non-BOLD data and a slope predicted by the echo times for BOLD data. 2/8

How do we define "good" fMRI data? Especially with resting state, there are circularity risks if we evaluate data quality as showing the networks we expect to see. Javier Gonzalez-Castillo (& me & others) developed pBOLD, a new metric that uses multi-echo info. www.biorxiv.org/content/10.6... 1/8

Picture of Alex Martin, National Institute of Mental Health

The Laboratory of Brain and Cognition at NIH is hosting a two-day symposium on 'Foundations and Frontiers in Cognitive Neuroscience' in honor of Dr. Alex Martin, to be held at NIH (with online videocast) on April 7th-8th, 2026. Register to attend online or in-person at: bit.ly/4bYlbxw

I recently saw a review with a (paraphrased) editorial comment "It looks like a reviewer used AI & the review is very nitpicky. Take or ignore the feedback." I'm glad the editors proactively commented. My guess is they thought the AI gave some semi-relevant feedback, but,as policy, this felt ad hoc.

That said, clinically targeted tasks (or a task battery) may require more time & customization than spontaneous fluctuations. That may be clinically less practical, which would make spontaneous fluctuations an inferior measure that's more practical to collect & thus (currently) more useful. 2/2

I think it depends on what clinical questions you are asking. My understanding is most clinical fMRI language localization work uses tasks with language. My assumption is, if you have a sense of a neuron-behavioral target, a task that drives that target will outperform spontaneous fluctuations. 1/2

I've been reading a great deal about current and potential impacts of AI on labor markets. Just published: a 2-page overview of frameworks for thinking about this: www.congress.gov/crs-product/...

I was sure the illusion would break at some length: can't keep grow forever, right?

Wrong. My brain hurts.

Probably ok unless you're arriving Sunday late afternoon / evening. @capitalweather.bsky.social updates often and usually has the most details and explanations on the range of possible storm outcomes.

Imaging the spontaneous flow of thought: Distinct periods of cognition contribute to dynamic functional connectivity during rest Brain functional connectivity (FC) changes have been measured across seconds using fMRI. This is true for both rest and task scenarios. Moreover, it i…

Not ICA, but this is using connectivity maps that are mapped to neurosynth labels. A very similar approach could be applied to ICA. This isn't hard to do, but I'd say all work like this should be interpreted with caution. www.sciencedirect.com/science/arti...

I also like building my own stuff, but if you're building something that you want others to possibly use, I wouldn't trust any AI to understand image header nuances yet. Chris Rorden knows them as well, if not better, than anyone & leads NiiVue, MRIcron, dcm2nii, +.

[AFNI Academy] APQC HTML: Overview YouTube video by AFNI Bootcamp

An example of niivue integration in AFNI's QC html report. youtu.be/hD9zTGMrAzQ?...

Or just use niivue.com with a really robust built-in API.

For context, this was a $50K+ full page ad in the NY Times.

From the neuro community on Reddit: Someone should really hire a neuroscientist at USC to look over their press releases... Explore this post and more from the neuro community

I just don't get why people keep relying on AI for stuff like this when one can get similar quality neuroscience images without AI. www.reddit.com/r/neuro/comm...

True story: it took me years to get a passport in my late 20s - at the time I did not have a single piece of identification with my name spelled correctly. All different misspellings. I eventually found a sympathetic clerk who was like 'I can see what happened here'.

Postdoc position to work on neuroimaging methods with @fmri-today.bsky.social (and me) fim.nimh.nih.gov/positions-av...

Over the years, I have written a few Jupyter/Rmd/Matlab notebooks that attempt to teach some statistical concepts, particularly in neuroimaging. You can find them here: www.mrc-cbu.cam.ac.uk/people/rik.h..., though I will say a bit more about each one in a number of posts over next few days.

papers.ssrn.com/sol3/papers....

I am recruiting a Postdoc to join my lab at UMN. If you or someone you know is interested in studying individual differences in brain and cognitive aging, check out the listing and my website in my bio and apply!

I appreciate RTs to help get the word out as well :)

With some trepidation, I'm putting this out into the world:
gershmanlab.com/textbook.html
It's a textbook called Computational Foundations of Cognitive Neuroscience, which I wrote for my class.

My hope is that this will be a living document, continuously improved as I get feedback.