Aleph captures highest-resolution non-invasive 3D brain images using ultrasound, open-sourcing the underlying datasets and code

The crazy hi-res image you’re seeing from groups like @alephneuro is an “ultrasound localization microscopy” or ULM image.

Groups like @PhysMedParis led by @TanterM have produced some insanely detailed images with this technique going back to ~2015.

What Aleph pulled off: imaging the blood vessels of a living human brain, in 3D, through an intact skull.

It’s an awesome achievement from the scrappy @alephneuro team - @_marleyx, @raffi_hotter, @lev & co 👏

A challenge I never thought I'd have:

Most of our renders of the brain were too beautiful to be used, and would make people think it was fake

So the real climb ahead: measuring brain function in humans that’s transcranial (i.e. non-invasvie), 3D + single-trial + real-time.

Each one alone is hard. E.g. single-trial functional imaging alone was the backbone of our Neuron 2021 paper & real-time was a major contribution in our Nature Neuroscience 2023 paper.

Together they're the frontier - the gap between a beautiful vascular map and someday decoding a thought non-invasively.

Tl;dr -- Aleph's result is mega-cool. A vial of microbubbles + a chip the size of a coin mapping a living human brain. It’s built on a decade-long revolution in ultrasound science and technology. And we’re just getting started 🔊🔊🔊

How ULM works:

Inject (FDA-approved) microbubbles into the blood. Each one lets you pin down a point far more precisely than the ultrasound wavelength itself. Track millions of them frame by frame, and the vessels emerge at micron scale.

This was born in Paris. The foundational result - Errico, Couture & @TanterM, Nature 2015 - turned microbubbles into super-resolution imaging. The @PhysMedParis / Institut Langevin folks built the field foundations from there.

(note: I put links to papers at the end of the thread)

Brain function is a different, harder problem - tracking blood-flow changes tied to neural activity.

Functional ULM *does* exist (Renaudin et al., Nature Methods 2022). It's currently working in anesthetized rats, skull open, averaging many blocks of trials to gain signal back. ~6.5µm, ~1s.

ULM is a structural image. It's a map of the pipes - anatomy and angiography. It's not (yet) mapping what the brain is *doing*.

That's the classic super-res tradeoff. To pin down each bubble you need them sparse, so you stack localizations over a long acquisition (e.g. ~4 min). You buy ~10µm resolution by spending temporal resolution. Great anatomy - but you can't watch the fast dynamics that comprise brain function.

A big enabling factor for all the advances you’re seeing in ultrasound is the hardware. Aleph built on the ultrasound-on-chip from @ButterflyNetInc - the same chip we used for sub-mm functional brain imaging at @ForestNeurotech.

One coin-sized chip, capable of whole heart ULM, whole brain functional imaging, and even whole body imaging @midjourney.

Getting through the human skull is hard - bone scatters and soaks up sound. But it's possible with contrast agents.

The first transcranial human ULM came from @CharlieDemene and the Paris group in 2021 (Nature BME).

Still in 2D, but absolutely landmark work.

3D-through-the-skull then showed up in animals - @JeanProvost's macaques through an intact skull, and sheep at Institut Langevin in 2024. Insane detail.

@_marleyx It’s super cool

@SumnerLN @slatestarcodex Relevant thread to one of your recent blog posts

@_marleyx Can you share what library you used to make this visual? Or is it custom

A great thread about our results!

For people who want to dig in even deeper, we also open sourced the data and a lot of the code

@levchizhov I can smell the tokens burning 🔥

@SumnerLN @levchizhov extremely cool

@_marleyx @alephneuro God is the best designer, architect, artist, logician, engineer all at the same time 100% overlapping

@SumnerLN @alephneuro Thank you for tracing this lineage! Super helpful