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

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'.

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

I'll also stress that I think this is nice data that prompts good discussion. From past experience, Nat Neuro likes publishing & promoting "Are we doing everything wrong?" manuscripts. Even if the text is more nuanced, I wished NN better supported & prepped the authors for the inevitable hyping. 5/5

Also, did you collect or analyze respiratory & cardiac traces, particularly for pCASL? I've found chest movement can create problematic artifacts and higher noise levels in the tag-control subtraction. I'm not sure there's a great correction, but it's worth understanding if there's an issue. 4/5

When originally skimming your paper, my first thoughts were this is plausible, but the deviations from previous assumptions were much larger than I'd expect given previous work in this area. Calculating and sharing empirical values for noise at each analysis step would help contextualize this. 3/5

Do you have empirical noise estimates from your existing data or phantom data that can be used to define more appropriate noise levels for @alexanderhuth.bsky.social's simulation? 2/5

@vavatin.bsky.social . Measures like CBF can be very noisy because they are the subtraction of two noisy time series, and you are also using models that multiply and divide data sources that can scale noise levels. 1/5

I suspected something similar, but thank you for creating a simulation and demonstration. I love papers that try to collect more quantitative information to push our understanding of MRI measurements, but we always need to grapple with measurement noise. The openly shared data do look interesting.

So I made a little simulation where there is no actual discordance, i.e. the underlying values from which the CBF, CBV, BOLD, and T2* are measured all move in lockstep. With zero noise this gives you zero discordant voxels (nice little diagonal line).

FWIW, this is my limited experience with pCASL. We tried a breath holding + paced breathing task and realized that the chest movement from paced breathing had a big effect on the tag-control calculation even without any tagging pulse. fim.nimh.nih.gov/assets/prese...

"...where the classic BOLD assumptions fail" -> "...where the classic assumptions about the relationship between the BOLD contrast and neural activity fail"

One thing I'm keeping in mind for my closer read is that my limited experience with pCASL had results that were VERY sensitive to chest movement. While skimming, I didn't see any mention of chest movement or cardiac traces. 2/2

I'm also trying to block of time to read this carefully & it definitely seems worth a close read. Edge cases where the classic BOLD assumptions fail don't surprise me, but this seems to be more widespread than I'd expect, and in contradiction to past work. 1/2

5