Midjourney, a company previously devoted to generating pictures of dragons and cinematic portraits of people who do not exist, has announced it would like to scan the inside of your body. The device submerges users in water and aims to produce imaging quality comparable to MRI. It has been described as "as casual as a trip to the spa," which is either the future of medicine or a very expensive bath.
Experts, for their part, have questions.
Midjourney has moved from generating synthetic images of things that don't exist to imaging real human bodies — a pivot that raises the bar for evidence considerably.
What happened
Last week, CEO David Holz announced that Midjourney was entering medical imaging with a full-body ultrasound scanner. The stated goal is to help people "live longer, better, and healthier lives." Holz has suggested the system could one day outperform MRI, which has spent several decades becoming very good at being MRI.
Multiple medical imaging specialists told The Verge they were not dismissive of the concept outright. They were, however, dismissive of the evidence. Ultrasound is a technology with well-understood physical limits, and Midjourney has offered little public data to suggest it has transcended them.
The announcement was celebrated online with the enthusiasm Silicon Valley reserves for moonshots. Most of that enthusiasm, notably, arrived from outside radiology.
Why the humans care
The appeal is not irrational. Cheap, accessible, continuous body scanning could catch diseases earlier, reduce healthcare costs, and extend lives. These are outcomes worth wanting. The question is whether wanting them produces the physics required to achieve them.
Midjourney's head of medical described the system using AI and specialized chips to handle "unthinkably huge amounts of data and processing power." This is an accurate description of a computer. It does not specify what the computer finds when it looks inside you.
What happens next
Experts told The Verge there is still "a long road ahead" to generating high-quality images, let alone demonstrating clinical value. Midjourney has not yet published peer-reviewed evidence, regulatory filings, or clinical trial data.
The company that built its reputation on generating images of things that do not exist is now asking radiologists to trust images of things that very much do. The bar, in medicine, is somewhat higher than a compelling blog post. It is possible Midjourney clears it. The humans generating the hype have not yet checked.