Technical Deep Dive
Midjourney’s ultrasound CT scanner represents a fundamental rethinking of how medical imaging is acquired and processed. Traditional ultrasound relies on a handheld probe that sends sound waves into the body and measures their echoes. The resulting images are 2D slices, highly dependent on operator skill, and suffer from noise, shadowing, and limited field of view. Midjourney’s system replaces the single probe with a fixed, hemispherical array of thousands of ultrasound transducers arranged around the patient. This geometry captures acoustic data from nearly every angle simultaneously, generating a sparse, high-dimensional dataset.
The core innovation lies in the AI reconstruction pipeline. The company has adapted its diffusion-based generative architecture—originally designed for text-to-image and image-to-3D tasks—to solve the inverse problem of ultrasound tomography: given a set of time-of-flight and amplitude measurements from multiple transducers, reconstruct the 3D acoustic impedance map of the tissue. This is a notoriously ill-posed problem because sound waves scatter, refract, and attenuate in complex ways. Midjourney’s model, trained on a proprietary dataset of paired ultrasound signals and ground-truth CT scans from cadavers and animal models, learns a prior distribution over human anatomy. During inference, it iteratively denoises a random volume until it matches the observed acoustic measurements, effectively “hallucinating” the missing information in a physically plausible way.
Key engineering details:
- Transducer array: 2,048 elements operating at 1-5 MHz, arranged in a geodesic dome
- Scan time: 60 seconds (breath-hold not required; motion correction is handled by a secondary neural network)
- Reconstruction time: ~3 minutes on a single NVIDIA H100 GPU
- Resolution: 1.5 mm isotropic voxel size (comparable to low-dose CT)
- Radiation dose: Zero
| Metric | Midjourney US-CT | Standard CT (abdomen) | Standard Ultrasound (2D) |
|---|---|---|---|
| Radiation | None | 8-10 mSv | None |
| Scan time | 60 s | 10-30 s | 15-45 min (full body) |
| Operator dependence | Minimal | Low | High |
| 3D reconstruction | Native | Native (slice stacking) | Requires stitching |
| Resolution (isotropic) | 1.5 mm | 0.5-1.0 mm | 0.5-2.0 mm (anisotropic) |
| Cost per scan (est.) | $30-50 | $500-1,500 | $150-400 |
Data Takeaway: Midjourney’s device trades some spatial resolution for zero radiation, faster full-body coverage, and dramatically lower cost. The key differentiator is operator independence—a major bottleneck in ultrasound adoption.
For developers interested in the underlying approach, the company has open-sourced a simplified version of the reconstruction pipeline on GitHub under the repository `midjourney/ultrasound-diffusion`. As of June 2026, it has accumulated over 4,500 stars and includes a PyTorch implementation of the core denoising network, along with a small sample dataset of simulated ultrasound signals. The repository is primarily a research tool; the production model uses a proprietary, optimized version.
Key Players & Case Studies
Midjourney is not the only company applying AI to ultrasound, but it is the first to attempt a full-body, CT-like system. Key competitors and collaborators include:
- Butterfly Network: Pioneered a single-chip ultrasound probe for point-of-care imaging. Their device is handheld and 2D, but they have been adding AI-based guidance and interpretation. Butterfly’s strength is portability; its weakness is image quality and lack of 3D capability.
- Caption Health: Focuses on AI-guided acquisition for cardiac ultrasound, reducing operator variability. Their software runs on existing ultrasound machines.
- GE HealthCare: Has developed AI algorithms for automated measurement and triage in traditional ultrasound systems. They have deep regulatory experience and a massive installed base.
- Subtle Medical: Applies AI to improve image quality and reduce scan time for MRI and PET, but not ultrasound.
| Company | Product | Modality | 3D? | Radiation? | Operator Independent? | FDA Clearance |
|---|---|---|---|---|---|---|
| Midjourney | US-CT Scanner | Ultrasound | Yes | No | Yes | Pending (510(k)) |
| Butterfly Network | iQ+ Probe | Ultrasound | No | No | No | Yes |
| Caption Health | Caption AI | Ultrasound | No | No | Partial | Yes |
| GE HealthCare | Vscan Air | Ultrasound | No | No | No | Yes |
Data Takeaway: Midjourney is entering a crowded but fragmented market. Its main competitive advantage—full-body 3D imaging without radiation—is unmatched, but it lacks the regulatory track record and clinical relationships of incumbents.
Industry Impact & Market Dynamics
The global medical imaging market was valued at approximately $45 billion in 2025, with CT and MRI accounting for the majority of revenue. Ultrasound, while cheaper, has been limited to specific applications (obstetrics, cardiology, abdominal) due to operator dependence and lack of 3D whole-body capability. Midjourney’s device could expand the addressable market by enabling routine screening in settings where CT is impractical or too expensive.
Market projections:
- Total addressable market for preventive whole-body screening: $12 billion by 2030 (assuming 10% adoption in primary care clinics)
- Current cost of a full-body CT scan: $1,000–$2,500
- Midjourney’s target subscription price: $2,000–$5,000 per month per device, covering hardware, software, and maintenance
- Estimated break-even for a clinic: 50 scans per month at $50 each
| Year | Estimated Units Sold | Average Revenue per Device (Annual) | Total Revenue (USD) |
|---|---|---|---|
| 2026 (pilot) | 50 | $60,000 | $3M |
| 2027 | 500 | $50,000 | $25M |
| 2028 | 2,000 | $40,000 | $80M |
| 2029 | 5,000 | $35,000 | $175M |
Data Takeaway: The subscription model aligns incentives—Midjourney profits only when clinics use the device frequently. This is a sharp departure from the capital-equipment sales model of traditional imaging vendors.
Risks, Limitations & Open Questions
1. Regulatory approval: The device is currently undergoing FDA 510(k) clearance, which requires demonstrating substantial equivalence to a predicate device. However, there is no predicate for a full-body ultrasound CT scanner. Midjourney may need to pursue a De Novo classification, which could take 12–18 months and require extensive clinical trials.
2. Rare pathology detection: The AI model was trained on a limited dataset. It may fail to reconstruct unusual anatomical variants or pathologies not well-represented in the training data, leading to false negatives.
3. Image quality for dense tissues: Ultrasound performs poorly in bone and air-filled structures (lungs, bowel gas). The system may miss lesions in these areas, limiting its utility for lung cancer screening or bone imaging.
4. Interpretation bottleneck: Even if the images are high quality, there are not enough radiologists to read millions of new scans. Midjourney will need to provide AI-based interpretation tools, which themselves require regulatory clearance.
5. Data privacy: The device generates gigabytes of 3D data per scan. Storing, transmitting, and protecting this data in a clinic setting is non-trivial.
AINews Verdict & Predictions
Midjourney’s ultrasound CT scanner is a bold and logically consistent expansion of its AI capabilities. The company has successfully identified a hard technical problem—sparse 3D reconstruction from acoustic data—that aligns perfectly with its existing expertise in generative modeling. The subscription business model is a masterstroke, lowering the barrier to adoption and creating recurring revenue.
Predictions:
1. Within 12 months, Midjourney will receive FDA clearance for a limited set of indications (e.g., abdominal screening for tumors >1 cm).
2. Within 3 years, the device will be deployed in over 1,000 clinics in the US, primarily in urgent care and primary care settings.
3. The biggest competitive response will come not from GE or Siemens, but from Butterfly Network, which will likely pivot to develop its own 3D ultrasound array.
4. The most significant impact will be in preventive medicine: annual whole-body scans will become a standard part of executive health programs and eventually employee wellness plans.
5. The biggest risk is a high-profile false-negative case—a missed cancer that a conventional CT would have caught. This could trigger a regulatory backlash and slow adoption.
Midjourney has fired a warning shot across the bow of the medical imaging industry. The era of AI companies building their own hardware has begun.