We checked how easy it is to match up people's brain maps after spatial normalisation. Turns out very! Maybe that surprising to many in the field - but my impression has long been that people are a bit in denial about deidentifying data. #neuroskyence #visionscience doi.org/10.1101/2024... - ThreadSky

sampendu.bsky.social • 25 days ago

We checked how easy it is to match up people's brain maps after spatial normalisation. Turns out very! Maybe that surprising to many in the field - but my impression has long been that people are a bit in denial about deidentifying data. #neuroskyence #visionscience

https://doi.org/10.1101/2024.12.19.629552

Comments

garilerma.bsky.social•15 hours ago

Interesting! we will test it with our new 10 subject/10 sessions dataset, anatomical coregistration washes out the functional coactivations, will inform back

sampendu.bsky.social•25 days ago

Anyone who knows my work knows I am a strong proponent of open data sharing. I do think though we should be a bit careful about how & what we share. Don't want to throw out the baby with the bathwater, but sensible custodianship or even just having some firmer access guidelines wouldn't go amiss.

sampendu.bsky.social•25 days ago

We had tried to publish a commentary about the implications of this in a journal - but unfortunately most of the reviewers felt this "lacks conceptual novelty" & that such procedures are already in place. I think this is delusional. That may be true where you work, but certainly not everywhere.

sampendu.bsky.social•25 days ago

I have decided (for myself so far anyway) that I don't have the energy - or the resources for that matter - to take this further as a project. So instead of fighting journals & reviewers about this I'll just leave up this preprint of the actual analysis.

kordinglab.bsky.social•24 days ago

So sad to hear - all revolutionary papers are mostly impossible to publish. At the risk of saying the obvious, there is a your name/ your paper joke in all of this that is probably not lost on you.

willjharrison.bsky.social•25 days ago

Interesting stuff, Sam. Can’t the maps still be de-identified, and unidentifiable, provided that none of the datasets link personal information to their maps?

sampendu.bsky.social•24 days ago

Sorry, I had to reread what you wrote. Yes, if there really is no further information linked in any of those studies that is true I suppose. But in many situations linking certain variables to the maps is crucial to answering the research question (e.g. about a health effect or drug use etc)...

sampendu.bsky.social•24 days ago

So unless you can guarantee that a person's map isn't linked to further personal information you cannot guarantee that someone further down the line will match up the data sets. And typically studies will link ages, gender, handedness, and possibly other demographic info to your physiological data.

remigau.bsky.social•25 days ago

Will check it out as I keep asking people: show me it can be done!
Though I have also heard people being stopped from sharing group level stat maps because they were told it could lead to individuals being reindentified. And that sounds wild to me.

sampendu.bsky.social•24 days ago

As in group averages? Yes that sounds wild to me, too. I would need a clear explanation for how that should even work. I can see how a particular group average might give you high confidence that someone is in a particular group (if variance is low enough) but not identify them individually.

benediktehinger.bsky.social•25 days ago

Cool stuff!

I wonder though, is datamatching really a problem if you need an fMRI session of the target person to do so?

Iirc according to GDPR's definition, this would still fall under anonymous data

sampendu.bsky.social•25 days ago

I'd say so. It's perhaps mostly a local-scale problem: Let's say you scanned someone & you find out they were scanned for another study where also some sensitive data was collected. It makes it fairly easy to match up participants & find out things that person might not want you to know...

sampendu.bsky.social•25 days ago

Yes, obviously this would be unethical. And yes, you could obviously do that if you already have access to the raw data (e.g. as a researcher or radiographer in an MRI centre). But the difference is that there -are- guidelines in place for that & you need specific access to those data to do it.

sampendu.bsky.social•25 days ago

I guess what concerns me about our simple result here is how you don't need specific access at all (as both data sets would be public) & people are being lulled into a false sense of security by telling them data are deidentified and/or masked by spatial normalised.

sampendu.bsky.social•25 days ago

Perhaps we don't need to change much, but before data like these are shared publicly we should be more explicit in our informed consent process to explain exactly how such data could potentially be matched up to similar data sets - including everything that those data are link to.

sampendu.bsky.social•25 days ago

In fact, I've started doing exactly that in my own information sheets. And according to Reviewer 2, such statements apparently exist in some other centres. But I do not believe they are the norm.

visscherlab.bsky.social•24 days ago

I'm not quite sure I agree -- anatomical data have exactly the same issue, unless I'm reading you wrong. That is, T1, or high res BOLD data can identify idiosyncratic patterns of folding, or other anatomy.

visscherlab.bsky.social•24 days ago

But you need to know those patterns to match them, so it's considered non-identifiable information. This is why we blur the face on any image. The face is identifying, but the brain isn't, because we don't see the brain...

Comments

Posting Rules

Reply