Midjourney 发布全身超声 CT 扫描仪与 Spa 计划

It’s a tough choice whether or not the buzzy Midjourney Medical launch today counts as AINews. Yes, Midjourney is one of the most significant and unique AI labs in the world. No, as David Holz was quick to point out, there’s not even any AI immediately present in the Scanner or Spa. But yes, ultrasound CT imaging obviously needs heavy AI assistance, and unlocks massive new applications downstream as BioHub’s CryoEMs did for ESM. And no, as Hacker News is quick to point out, there are a lot of unsolved questions as to how ready or useful this scanner really is.And yes, that was me in the livestream, which we are transcribing to save you 2 hours. Overall the vibe was electric and inspiring; I sat next to Robert Scoble, who was there at the original iPhone and Tesla (and Google Glass) launches, who agreed that this launch was comparable in ambition; and next to Tanishq Abraham, who, by sheer coincidence, just so happened to tweet out this Nature paper on ultrasound CT:On to the facts you must know.Facts / AnnouncementsMidjourney announced a medical imaging project, calling it the Midjourney Scanner.The device is described as a full-body ultrasonic CT / full-body ultrasound system.David Holz framed it as the “first new whole-body medical imaging modality in 50 years.”The scanner uses ultrasound rather than MRI, X-ray, or CT radiation.The system involves:8,960 transducers per chip/system40 systems arranged in a ring358,000 ultrasonic elements totalA 70 cm diameter ringWaves traveling through water at about 1,481 m/sData capture around 17 GB/sAround 40 GB of data per body sliceReconstruction using 21 serversClaimed 2 PFLOPS computeClaimed 806 TB raw dataLift movement at 4 cm/sGoal of several hundred slices in 60 secondsClaimed resolution of internal tissue details down to about 0.5 mmCurrent imagery shown included:Real body slicesComparisons with MRI, especially thigh/muscle boundary imageryUltrasonic phantom imagesSegmentations of organs and biological structuresThe current system is Gen 1 / prototype, not the finished consumer product.Midjourney says it is not yet using AI for the shown images.The team size working on the device is about nine people.About a dozen people have been scanned so far.Current scans can take around 20 minutes, because the system is still bottlenecked by bandwidth, algorithms, DSP, and prototype data-transfer infrastructure.Midjourney also announced the Midjourney Spa:First location: around Union Square, San FranciscoAbout 25,000 sq ftFour floorsHot tubs, saunas, cold plunges, gym, and other spa amenitiesAround 9–10 scannersLease signed and designs underwayDesigned by architects associated with major spa projects such as Blue LagoonTarget opening: end of 2027Midjourney says it is self-funded, has no investors, and can fund the first spa itself.The company says it has started discussions with the FDA.Initial regulatory/commercial path is likely around body composition, because that is considered easier.Jobs and more info are expected at midjourney.com/medical.Projections / Goals / Claims About the FutureMidjourney’s stated long-term goal is a fleet of 50,000 scanners.Claimed goal: enable up to 1 billion scans per month, enough to bring full-body imaging to everyone.Holz suggested that fewer than a dozen such machines, operating at full speed, could perform more full-body scans than all MRI machines on Earth combined.The company expects:Gen 2 scanner by the end of 2026Gen 3 scanner with custom silicon laterFuture systems to become much more advanced through custom chips, AI, physics simulations, and better compute infrastructureHolz projected that the scanner could eventually support:Frequent personal health trackingDaily/weekly/monthly body scansPreventive medicineDetection of “weird” changes in the bodyBody composition trackingDoctor-facing reviewAI-assisted first-pass analysisPotentially thousands of diagnosesEventually, some therapeutic usesHe speculated that preventive imaging could reduce healthcare costs substantially, possibly by catching disease earlier.He suggested the scanner could become hundreds or thousands of times cheaper than MRI on a per-scan basis, because the machine is cheaper and faster.He said the marginal cost of a scan could be effectively zero, though the actual business model will involve spa/facility economics.Possible pricing models mentioned:Spa membershipsWalk-in scansScan-only pricingSpa-only pricingSome broader pricing matrixThe first spa is intended as a learning lab for usage patterns:Do people want full spa + scan?Gym + scan?Quick scan and leave?Daily, weekly, monthly, or annual scanning?Holz estimated scaling to thousands of spas could require around $20B in upfront capex.He speculated the facilities might pay themselves back quickly, even mentioning six months, but explicitly caveated uncertainty.Therapeutic uses were described as long-term, not day-one:Tendon/muscle healingFocused ultrasoundPotential incisionless proceduresPossibly cancer tissue destruction at a distance, but explicitly not near-term.Opinions / Vision / FramingHolz framed Midjourney as a community-supported research lab, not a normal VC-backed startup.He repeatedly emphasized that Midjourney’s image-generation revenue gives it freedom to fund ambitious R&D.The scanner was presented as part of a broader mission around positive human futures, not just creativity tools.He described the desired experience as:“As powerful as an MRI”“As casual as a trip to the spa”He does not want scanning to feel like a doctor’s office.He wants the spa to be desirable even without the scanner.He personally wants frequent health feedback so everyday diet/exercise choices become measurably visible.He sees the scanner as a possible new pillar of AI-enabled healthcare: AI needs fast, rich, cheap bodily data.He argued that the future is not only about AI models but about new infrastructure that lets AI reason over the physical body.He sees ultrasound as the right modality because it can be fast, safe, dense, and data-rich.He appears especially excited about longitudinal, high-frequency, sub-millimeter differential tracking: not just “one scan,” but changes over time.He positioned this as “day one of MRI” for full-body ultrasonic CT: early images may look rough, but the modality could improve dramatically.Reasons / RationaleWhy ultrasound instead of MRI?MRI is hard to make both fast and high-quality.MRI scans are unpleasant: tubes, long sessions, loud sounds.Ultrasound can push more energy through the body safely.Ultrasound has no ionizing radiation.Ultrasound can be repeated often.Ultrasound is already widely used medically, making some regulatory paths easier.Why water immersion?Sound travels through water much faster and more effectively than through air.Water coupling enables whole-body ultrasound propagation.The design requires the user to get wet, hence the spa concept.Why vertical up/down scanner design?Optimized for throughput.Easier than having users lie in and climb out of tubs.“Down and up” scanning supports faster repeated use.Why build a spa first?To learn real-world operations.To test throughput.To understand consumer behavior and willingness to use scans.To refine pricing/business model.To gather data.To create a replicable template before scaling globally.Why Midjourney can attempt this?Existing image business generates revenue.No investors means fewer constraints.Midjourney already has compute infrastructure.The company has skills across AI, imaging, sensors, visualization, and systems engineering.Holz has prior hardware experience from Leap Motion.Why not launch as pure medical device immediately?FDA/regulatory path is complex.Some use cases are easier than others.Body composition is an easier initial wedge.Diagnostic and therapeutic claims require staged approval.Why cloud processing?Raw scanner data is enormous.On-site compute can handle streaming/compression.Midjourney’s large server clusters can process heavy reconstruction worklo