Simon Eickhoff

Warum ist es so schwer, KI fair zu gestalten? Mein Vortrag (morgen 7.10. um 14 Uhr) in Essen oder online per live-stream: lnkd.in/e9XiAS3h

In a meta-analysis of 210 biomedical AI studies that statistically compared models under cross-validation, 97% used invalid statistical tests. Here's our new preprint doi.org/10.64898/202... led by @tianchu.bsky.social @hetuli.bsky.social @shaoshiz.bsky.social @nichols.bsky.social 1/N

Hello Bluesky! Below this post, you will find all my very interesting "new paper" threads. 🤓 Note that some of these are much older than the timestamp because I rescued them from Twitter to make them accessible for everyone.

Despite decades of research there are still no reliable biomarkers for any of the major psychiatric diagnoses. In this now published perspective in Molecular Psychiatry we consolidate evidence from genetics, exposome and neuroimaging research on the potential reasons underlying these failures.

It's important to note that when discussing brain structure-function relationships across scales, that the definition of function changes with scale. Keep it straight! @sbe.bsky.social link.springer.com/article/10.1...

🚨 New preprint alert! 🚨 Transdiagnostic latent factor models of psychopathology are widely assumed to improve brain-behaviour associations. So we decided to test this directly and found that they don't. A short 🧵 Link: www.biorxiv.org/content/10.6... 1/10

👉Up next, we are excited to welcome @sbe.bsky.social for our next CCP Speaker Series talk! Join us THIS FRIDAY, December 5th, at 1pm EST RSVP HERE for more details: docs.google.com/forms/d/e/1F...

✨ New preprint ✨ Introducing cuBNM, a toolbox we developed to run brain simulations *much* more efficiently on GPUs 🎉🚀 Preprint: www.biorxiv.org/content/10.1... Documentation: cubnm.readthedocs.io/en/stable/ GitHub: github.com/amnsbr/cubnm 1/n 🧵

✨ SHINY NEW PAPER ✨ Sleep-disordered breathing is associated with higher Aβ burden, lower rFDG, and worse cognition in Alzheimer's Disease. @sbe.bsky.social www.sciencedirect.com/science/article/pii/S0197458025001034

Excited to see this work published in @natcomms.nature.com! 🧠📊 In-vivo evidence of E/I alterations underlying functional changes in individuals with autism in the large ABIDE1 dataset AND replication in the independent ABIDE2 dataset.

New preprint from stellar IRTG PhD student Amelie Rauland + team on white matter bundle reconstruction! Shows that WM bundles can be reliably extracted from simple 32-direction dMRI & features predict cognition - huge potential for legacy and clinical data. Thread 👇 www.biorxiv.org/cgi/content/...

The end of a PhD journey 🎓 What a ride - full of lovely people, exciting science, coding tunnels, coffee-brainstorms & amazing conferences! Whole-hearted thank you to all who shared the journey & celebrated with me - especially @sofievalk.bsky.social, @sbe.bsky.social &@mps-cognition.bsky.social 💐🥳❤️

1/11 Excited to share our @Naturestudy led by @leonooi.bsky.social @csabaorban.bsky.social @shaoshiz.bsky.social AI performance is known to scale with logarithm of sample size (Kaplan 2020), but in many domains, sample size can be # participants or # measurements... doi.org/10.1038/s415...

Amelie Rauland (all star via IRTG student w/ @sbe.bsky.social ) will be presenting "Benchmarking Reconstruction Methods for Bundle Segmentation in Single-Shell dMRI" on Friday and Saturday, Poster #1283

#ABCD-J #Studienhighlights: Eine kürzlich erschienene Studie zeigt mittels #JTrack Social Unterschiede in Kommunikationsmustern zwischen neurotypischen Erwachsenen und Erwachsenen mit #Autismus-Spektrum-Störung auf. Neugierig geworden? Das ganze Highlight gibt es hier: www.abcd-j.de/de/showcases...

🧠⚖️📉 How does the cortical excitation-inhibition ratio mature during adolescence? We asked this in our new paper just out in #ScienceAdvances ✨ “Adolescent maturation of cortical excitation-inhibition ratio based on individualized biophysical network modeling” 📄 www.science.org/doi/full/10.... 🧵⤵️

⭐ new paper Maturation of cortical excitation-inhibition (EI) in adolescence now in Science Advances 💥 A true tour de force of multimodal MRI, connectomics & GPU-powered 🤖!! biophysical modelling by hiball superheroes 🦸‍♂️🦸‍♀️🦸‍♂️ @amnsbr.bsky.social @sofievalk.bsky.social @sbe.bsky.social

Our #metaanalysis (n = 231) in JAMAPsych reveals consistent brain abnormalities across #sleep disorders involving affective and cognitive hubs in sgACC, amygdala/hippocampus Short-term sleep deprivation shows a distinct picture, affecting the thalamus jamanetwork.com/journals/jam...

It's finally here! Use the Network Correspondence Toolbox to help contextualize your neuroimaging findings 🧠

Always thought we should use connectivity blueprints to investigate differences between human, chimpanzee, and macaque brains? www.jneurosci.org/content/earl... For Katherine Bryant’s latest masterpiece.

Now out in NPJ Digital Medicine our manuscript on real-world evidence for altered communication preferences in autism: nature.com/articles/s41... @sbe.bsky.social @leoschilbach.bsky.social

A new #Neurosalience episode is out! 🧠 From Big Data to Biomarkers with Peter Bandettini @fmri-today.bsky.social and Simon Eickhoff @sbe.bsky.social youtu.be/z91SiENsgF8

Dazu integrieren wir – das Institut für Neurowissenschaften und Medizin: Gehirn und Verhalten (INM-7) @fz-juelich.de, und unsere klinischen Partner an den Unikliniken Aachen, Bonn, Köln, und Düsseldorf – verschiedene technische Lösungen zu einer digitalen Plattform und vernetzen uns.

Thanks to everyone's feedback, we have updated our calculator to optimize sample size N & scan time T for fMRI studies: leonoqr.github.io/ORSP_Calcula... The first new feature is that users can explore how different N & T leads to different accuracy, e.g., N=1000 & T=30min => 81% max accuracy. 🧵

The limiting factor for MRI based prediction of behavioral traits may be neither the imaging nor the pipeline or learning algorith but rather the reliability of the target phenotypes Thought provoking work my Martin Gell et al: www.nature.com/articles/s41...

This is what happens if preprints are treated as „scientific literature“ Don’t get me wrong, I am all for preprinting. But unfortunately, it is often ignored that anybody can preprint anything

We created a #neuroinformagical (obvs) starter pack! Let us know who else to add! #openscience #datasharing #neuroskyence 🧪 go.bsky.app/LRqnaMP

woo hoo - fantastic new work from ‪@wanb.bsky.social‬ @sofievalk.bsky.social @sbe.bsky.social & many more on microstructural asymmetry now out in Nature Comms ⤵️

Useful reminders from @sbe.bsky.social about #neuroimaging analysis #reproducibility at #OHBM2024: - Different atlases lead to different connectivity matrices. - The target *massively* matters. - ML algorithms just *love* shortcuts.

Lots to enjoy in this new paper by Sam Vickery @sbe.bsky.social and others. Brain areas in humans that show the greatest expansion from chimp -> human are the areas that are most susceptible to to the effects of aging www.science.org/doi/10.1126/...

We are looking for a community manager and development coordinator for DataLad !! The full description with contact info is at: files.inm7.de/mih/position...

Spread the word 📣 Announcing the first BrainHack at the DGKN meeting next month !! After being a staple at OHBM, this will be the first hackroom at a German (clinical) neuroscience meeting Enjoy the spirit of open science with tutorials, hackathons and an unconference 🔜 Join us in Frankfurt

Somehow it seems that computational models were immune to the current sample-size worries in #neuroimaging Turns out they are not 😱 DCM models and connections need 100+ subjects to give stable results New work by Alexander Silchenko & Felix Hoffstaedter: onlinelibrary.wiley.com/doi/10.1002/...

That is a really important result. These models are being run on an n of 12 or 20.

Somehow it seems that computational models were immune to the current sample-size worries in #neuroimaging Turns out they are not 😱 DCM models and connections need 100+ subjects to give stable results New work by Alexander Silchenko & Felix Hoffstaedter: onlinelibrary.wiley.com/doi/10.1002/...

Finally in the public 🚀 JuLearn: an easy-to-use library for leakage-free evaluation and inspection of ML models JuLearn is an open-source Python library simplifying the entry into the ML world -> design & evaluate ML pipelines without encountering common pitfalls arxiv.org/abs/2310.12568

Fantastic but very worrisome work by Sami Hamdan: Confound removal in machine learning can itself lead to leakage -> After information about the target from the features using standard approaches you can STILL predict it well academic.oup.com/gigascience/... #neuroskyence #MLSky #neuroimaging

Finally out 🎉 - our meta-analysis of 403 neuroimaging studies on language (3/3) Quite some functional specialization in subcortical areas, too. This and many more details can be found here: psycnet.apa.org/fulltext/202... Paper is open access !!

Finally out 🎉 - our meta-analysis of 403 neuroimaging studies on language (2/3) We ended up discovering truly subdomain-specific brain areas in spite of substantial covergence Functional specialization was most pronounced in the cortex but we also found considerable cerebellar effects

Just wondering, am I the only person living (and loving) inbox-zero for years? I use my inbox as a todo list and make extensive use of „defer / put back“. Works amazing and each time I am back to empty it feels rewarding

A main goal of our institute is to provide open software & resources - DataLad (data-management) - JTrack (mHealth / wearable suite) - JuLearn (user-friendly ML library) - ANIMA (meta-analyses results) - JuSpace (MRI-PET integration) - many more... Full catalogue: www.fz-juelich.de/en/inm/inm-7...