Something’s off with Midjourney’s pivot to body scanners

Experts say the AI startup’s bold statements need proof.

Experts say the AI startup’s bold statements need proof.

Last week, Midjourney, an AI startup best known for its image generator, made an unusual pivot: medical imaging.

The company announced a futuristic ultrasound scanner that would dunk users into a vat of water and, hopefully, produce “something as powerful as MRI” yet “as casual as a trip to the spa.” Midjourney says the goal is to help people live longer, better, and healthier lives. CEO David Holz has suggested the system could one day be better than MRI. Experts are skeptical. While several medical imaging specialists told The Verge they were not dismissive of the idea outright, they said Midjourney has shown little public evidence to substantiate its goals —especially for a technology that has been around for decades and has well-understood limits.

To call this a left-field move is an understatement. Midjourney is shifting from generating synthetic images online into the high-stakes, tightly regulated world of medicine — and definitely not-so-tightly regulated realm of wellness. It is not immediately clear how the venture relates to Midjourney’s existing AI business model, and AI is hardly mentioned in the blog post laying out the company’s plan. Midjourney’s head of medical Tom Calloway told The Verge the scanner uses AI and specialized chips to handle the “unthinkably huge amounts of data and processing power” required to “execute a scan.” Calloway said that AI is also used “to enable lossless compression and dramatically speed up processing.”

Online, the announcement was celebrated as exactly the kind of disruptive moonshot Silicon Valley loves, with some hailing it as the future of medicine and a path for cheap, constant monitoring that could and prolong lives (while making Midjourney a boatload of cash). Much, though not all, of that enthusiasm came from outside medicine and radiology.

Radiologists, clinicians, and imaging experts who spoke to The Verge were less awestruck. Several said the concept is genuinely exciting and maybe even plausible. But they also said the idea is not as novel as Midjourney suggests, raised basic questions about its execution, and all asked the same crucial question of the company: Where is the proof?

That question looms over every part of Midjourney’s vision: its comparisons to MRI, the number of scans they say they can perform, the design of its scanner, the fidelity of its images, and the hope that frequent imaging could save and prolong lives.

The machine at the center of “Midjourney Medical” is an ultrasound scanner that would lower users into a vat of water while they stand on a platform. Once submerged, a ring of underwater sensors and scanners would send sound waves into the body and capture the echoes that bounce back to generate internal images. The company likens the process to dolphin echolocation and says the goal is for it “to take no more than 60 seconds.” Typical ultrasounds can take around 30 minutes or more to perform, though length depends on the nature of the test and body part being scanned. MRI scans — which use powerful magnets and radiowaves to generate detailed images of the body’s internal structures — often take longer and can be uncomfortable, requiring patients to lie still inside a narrow, noisy tube. Other ways of generating detailed internal images, such as CT scans, often use ionizing radiation, which makes unnecessary or repeated scans a safety concern.

Midjourney says the scanner is meant to give people more data about their bodies to help them make better decisions about their health. It compares the process and images to MRI scans and even quotes statistics claiming that “the world could avoid 30% of all deaths and 50% of all healthcare costs” if only there were enough early imaging.

But despite that medical-sounding language, Midjourney is not initially framing the scanner as a diagnostic medical device, citing the high FDA clearance and clinical trials required to market the scanner for medical use. (Though Midjourney says it plans to expand into medical applications later.) The company describes it instead as a way to give people more information about their bodies and plans to embed the machines in spas, where scans will become “a side-effect” of visits. Concept images show luxurious golden rooms and pools of water, about as far from a doctor’s office as imaginable.

“The technology is super cool,” said Venkatesh Murthy, a professor of preventive cardiology, internal medicine, and radiology at the University of Michigan Medical School. He told The Verge he is excited to see teams building working prototypes of technology like this, an idea he said has been around for a long time. But Murthy cautioned that there is still “a long road ahead to generating high-quality images and then to understand the clinical value and demonstrate net benefit to patients.” He said many of Midjourney’s “claims about resolution are clearly theoretical” and suggestions it could be an “MRI equivalent are completely unsupported.” The images the company has shown so far, he said, “are decidedly low-res.” (“All of these images come together to cover a 3D map of your body, down to a fraction of a millimeter, that looks a lot like today’s MRIs,” the Midjourney blog post reads.)

Mark Anastasio, a professor of imaging sciences at Washington University in St. Louis, described Midjourney’s entry into the medical imaging space as an “exciting development” and said the idea of the ultrasound scanner was certainly plausible. Ultrasound has a long history as a diagnostic tool in medicine, he noted, and researchers have recently demonstrated whole-body prototype systems using related approaches. However, he said it would need validating before making any medical or diagnostic claims, adding that there “is no current evidence” that detailed ultrasound scans like this could be comparable to MRI. Matthew Davenport, a professor of radiology at the University of Michigan Medical School, had a blunter assessment. He said Midjourney’s published images were “interesting” and that he could see a market for body imaging. But the company’s “claims are wildly unsubstantiated,” he told The Verge — “perhaps the most grandiose” he has seen.

Following up on comparisons to CT and MRI scans, Calloway told The Verge: “Every modality has its strengths and weaknesses. MRIs are powerful but expensive and slow. CT scanners are excellent for things like bone and lung imaging but expose patients to ionizing radiation. The first-generation scanner is fast, affordable, radiation-free, and great at imaging soft tissues like muscle and fat for body composition — but it won’t be a replacement for either of these technologies.”

Scott Reeder, a professor of radiology at University of Wisconsin School of Medicine and Public Health, also described the concept as “innovative,” but said it “is not a proven technology,” with little evidence on Midjourney’s website to support the company’s suggestion it could be used clinically. He said it is “a stretch” to compare ultrasound scans like this to MRI or CT scans, which provide different sorts of information and have different limitations. Ultrasound relies on sound waves moving through the body, Reeder explained, meaning air and bone can be major barriers. That could make it harder to image parts of the body blocked by or involving these, he said, including the gut and organs inside the pelvis, particularly without a technician to maneuver probes into a better position. Heads would also be tricky, though Midjourney’s scanner does not fully submerge users.

William Morrison, a professor of radiology at Thomas Jefferson University, was the most skeptical about Midjourney’s proposal. He called the whole thing a “vibe-based rollout” supported by images that “are markedly limited compared to existing technology such as CT and MRI.” Morrison described ultrasound as a “very old technology” and said the water bath approach had been largely abandoned because of the physical limits of sound waves making such an approach difficult to implement.

The water itself would also be a problem. Morrison said it would need to be “completely pure, without bubbles or dirt” to produce good images, something that would be expensive to maintain and likely require changing between customers. Dirt and hair could trap air or deflect sound waves as well, Morrison said, meaning to get the best images customers would probably “have to be scrubbed and completely shaven.” (In response to concerns about water purity and bubbles, Calloway told The Verge “the bubbles are removed from the water with a large degasser located under the tank.”)

Morrison also raised the same basic physics problem as Reeder: Sound waves struggle to penetrate the body deeply enough and can be blocked or distorted by air and bone. Fat can also quickly attenuate ultrasound waves, he said, and give signals more tissue to travel through, raising questions about how useful such a tool would be for people with larger bodies. “I noticed that the images provided by the company are acquired on very thin people,” he said, suggesting that this was a deliberate choice and that images of larger patients would be of lower quality. Calloway said this is a challenge CT and MRI scans also run into, adding that Midjourney is “actively working with our partners” to address the issue.

In line with how many AI health products are marketed, Midjourney has conveniently given itself room to sidestep many of these concerns. By launching the scanner as a wellness product and not a diagnostic medical device, and by only offering “detailed body composition maps,” Midjourney is hinting at future parity with MRI images, longer lives, and better health without actually having to prove it. The company says it plans to submit “regular test results” to the FDA to expand authorized use cases.

Murthy agreed that “body composition is plausible” as a use case, but noted “much of the messaging [from Midjourney] isn’t about body composition but about cancer screening and overall lifespan prolongation.” Body composition can already be measured with available technology, he said, adding that some weighing scales “will give you similar measurements with slightly less accuracy.”

It is also unclear how Midjourney will explain this difference to customers, or how clearly it will spell out the difference between body composition data and medical information. Morrison said he would expect the company to use some kind of disclaimer to protect itself if the scanner missed something important. Midjourney declined to provide specifics, but Calloway said the company is “being highly intentional” about the scanner’s interface.

“We’re strictly bounding the system to general wellness, building beautiful, detailed 3D maps of body composition,” he said, adding that Midjourney has “confirmed this classification with the FDA and are operating squarely in that lane.” In the future, he said more capabilities will be added in partnership with regulators and medical experts. Midjourney’s blog already outlines multiple planned generations of the scanner, with a third-generation planned for 2028, when it says the technology will get “serious” and “image quality and scan times will be night-and-day.” By 2031, Midjourney says it wants to have more than 50,000 scanners worldwide, enough, it says, to give monthly scans to a billion people. Midjourney says its first spa will open in the “heart of San Francisco” in 2027.

Reeder said it “would be concerning if patients replaced proven screening technologies with this technology” — for example if someone skipped a mammogram or colonoscopy because they believed a Midjourney scan had already checked them sufficiently. But even if the technology is eventually proved capable of detecting disease in a way that allows Midjourney to market it for medical use, mass scanning would not automatically be a good idea. Elective full-body imaging is already contentious, a debate Reeder and Davenport have both written about in a Journal of the American Medical Association article on MRI screening last month.

In the end, the concern over Midjourney’s pivot into medicine comes back to proof. The idea of building a machine that improves peoples’ health is hard to argue with. It underscores why proof is so important — laudable ideas are what makes the vision of companies like Theranos so compelling. It is why companies entering the field should have the evidence in hand before promising a transformation so momentous that it sounds almost too good to be true.

“The interest in improving health through imaging is wonderful,” Davenport said. “The race to market with unproven claims that almost certainly will not prove true is ethically problematic.” Morrison’s appraisal was harsher: The whole move, he said, has the feel of an ad campaign. “It makes me think that this may be more of a grift than a pivot.”

• Robert Hart