I'm getting a lot of questions about @midjourney's new scanner, so here's my take. This article is not meant to be a critique of their technology; I just want to introduce some nuance in the public conversation, as someone who has spent the past decade in medical imaging. I also consulted on the topic with my Chief Medical Scientist at @ezrainc / @function, Dr. Dan Sodickson (@DanielSodickson), a world-renowned scientist who invented parallel imaging, a technology that's used in every modern MRI, and who recently wrote a book on the future of imaging.
TLDR: I'm delighted and bemused in equal measure by Midjourney’s scanner and claims. They've gotten some things right and hand-waved others with a real dose of naivety.
First, a bit of history. Water immersion isn't futuristic, it's where ultrasound started. The first 2D scanners in the 1950s submerged patients in a cattle tank, then literally an upturned B-29 gun turret, weighted down so they wouldn't float. Radiology abandoned immersion in 1958, when contact gel scanning arrived, because water immersion was impractical. So this is a revival of the thing the field walked away from 70 years ago.
The deeper issue is physics. Ultrasound doesn't travel through bone or air (hence the water immersion). And it turns out we have quite a bit of both inside us. It's not that ultrasound can't see bone – bone surfaces actually show up bright and clear – it's that it can't see through bone or air. So the lungs, and anything tucked behind your ribs, spine, or bowel gas, are effectively off-limits. Sound energy halves every 360cm in water, but every ~2mm in bone and ~0.6mm in air. Your torso is ribs, spine and pelvis wrapped around lungs and gas-filled bowel. There are windows between bony structures that allow good visualization, but crisp whole-body visualization is close to the worst-case target for sound-based imaging. Notice that the Midjourney image stacks don’t include the brain - an important structure entirely encased in bone!
The tell on the bone / air limitation is in the one FDA-approved product that already works using water-immersion ultrasound technology. A ring array with tissue suspended in warm water – Delphinus SoftVue – is FDA-approved for breast imaging. It took them ~15 years to get it through the FDA, and it images the breast only. They chose the breasts because it’s the one region that's pure soft tissue, with no bone or air in the path.
Basically, a single whole-body snapshot using this ultrasound machine will be a diagnostic mess: whole regions shadowed out, uncertain signal, variable penetration, poor signal to noise and contrast to noise ratios across the board.
Then there are the claims. "Superior to MRI, 100x faster" is an eye-popping exaggeration at this stage. For one, what they’re actually shipping next year is a body-composition map (fat, muscle, water) which competes with a DEXA scan or an InBody machine, not with an MRI. It's not disease detection.
The 60-second scan is also a target, not today's reality: current Midjourney scans reportedly take ~20 minutes, and only about a dozen people have been scanned so far. And no radiologist has seen the resolution (at least not in any publicly shared materials), plus Holz himself concedes MRI is still better in some respects. You can't know whether these scans are diagnostically useful without extensive validation in diseased patients, cross-checked against a proven modality like MRI.
There's also the regulatory reality. Nothing they're launching needs FDA clearance, because a wellness body-composition map isn't a diagnostic device. But everything that would make this an actual MRI rival does. There's no existing predicate for a whole-body diagnostic ultrasound, which means a De Novo pathway, robust prospective clinical studies, and – realistically – a couple of years at the earliest before any diagnostic claim clears.
Oh, and for an AI lab, the kicker: Holz says they're "not even using any AI in this yet." It's hardware and signal processing – even the ultrasound chips at the core are licensed from Butterfly Network. The software on top only segments and labels the output images.
Now, I don't mean to suggest there's no use here. It's a clever revisitation of old technology, paired with genuinely interesting software to stitch the slices together and provide segmentation. But without a lot more validation, it's not yet… revolutionary.
All this said, the body-composition piece isn't nothing. A Caltech group (Lihong Wang's lab) published whole-body cross-sectional ultrasound tomography in living people back in 2023, and showed it's genuinely good at mapping fat and muscle around the whole body, with no radiation and no compression, catching adipose that calipers and even DEXA miss. As a body-composition and tissue-tracking tool, this approach has real, peer-reviewed validation.
The same Caltech work points to where this scanner actually shines, and it's not as an MRI-killer. It's longitudinal tracking: frequent, cheap, safe scans to watch broad structural change over weeks and months, the kind of monitoring MRI is currently too slow and expensive to do repeatedly (but not for long). The paper above even floats (ha!) wellness centers as a natural home for it.
So Midjourney's instinct is… sound: build a fast, cheap, radiation-free scanner. But the revolution isn't the snapshot; it's scanning longitudinally to detect change, fusing it with other modalities like MRI, and putting it in homes rather than spas. We’ve spent a lot of time thinking about this at Ezra / Function, and Dr. Sodickson even wrote a whole chapter about it in his book (see Chapter 11 if you want to geek out).
Overall, kudos to @DavidSHolz and his team for trying something truly (old but) new. In my opinion they've front-run the validation and oversold the destination, but the direction is real. Excited to see how far they take it!
