Midjourney Medical
Midjourney, best known for AI image generation, has unveiled an ambitious plan to build full‑body “ultrasonic CT” scanners deployed in spa-like centers, aiming for up to a billion low-cost scans per month and framing this as a path to early disease detection and massive medical datasets. Commenters are sharply divided: some see a bold, potentially transformative bet on non‑ionizing, high-throughput imaging and longitudinal health data, while many doctors, engineers, and statisticians warn about physics limits of ultrasound, false positives, overdiagnosis, regulatory hurdles, and “Theranos-like” hype, especially given the spa/consumer positioning and lack of published clinical evidence. Privacy, data ownership, and the mismatch between generating vast health data and having the medical infrastructure or evidence base to use it safely are recurring concerns.
Technical concept & feasibility
- Device is described as “full-body ultrasonic CT” using hundreds of thousands of transducers in water to reconstruct a 3D scan in ~60 seconds.
- Some with imaging backgrounds say USCT is real, used in research and breast imaging; submersion improves coupling (water–skin interface).
- Radiologists and imaging engineers note current sample images look low-detail compared to conventional ultrasound, CT, and MRI and doubt claims of “MRI-level or better” resolution, especially for deep structures, lungs (air), and brain (skull blocks sound).
- Others question data-rate claims (petabytes per scan) as marketing spin; actual pipelines would likely downsample heavily on FPGAs/compute nodes.
Medical value, false positives & overdiagnosis
- Many posters argue mass full‑body scanning of asymptomatic people is likely harmful: high false‑positive rates, “incidentalomas,” cascades of follow‑up tests, biopsies, surgery, anxiety, and system overload.
- Bayes’ theorem is repeatedly invoked: for low-prevalence diseases, even highly accurate tests produce mostly false positives.
- Counter‑view: cheap, frequent longitudinal scans (trends over time vs. one‑off snapshots) plus better models might eventually distinguish benign quirks from dangerous changes and radically advance medical understanding.
Regulation, liability & “spa” positioning
- Strong concern that marketing as a spa/wellness service is an FDA/regulatory workaround; skeptics say it should first prove clinical utility in hospitals and trials, not med‑spas.
- Questions about who reads scans, who is liable for missed findings, and whether non‑diagnostic “body composition maps” will still be interpreted quasi‑medically by users and some doctors.
Scale, economics & target users
- The stated goal (50k scanners, capacity for 1B scans/month) is widely viewed as mathematically and logistically unrealistic given scan time, cleaning, staffing, and infrastructure.
- Many think early adopters will be wealthy biohackers, not “a billion people,” and compare it to executive physicals and other luxury screening services.
Data, privacy & AI
- Some see the main play as building a massive unlabeled imaging dataset to train health models (analogous to LLMs on text).
- Others are uneasy about a generative‑image company owning full‑body 3D medical data, with little mention of privacy, governance, or openness.
Overall sentiment
- Mixed but skewed skeptical: admiration for ambition and hardware, but strong Theranos comparisons, concerns about hype, lack of clinical evidence, and the risk of widespread over‑diagnosis.