Tech

Midjourney Medical pivots from cat images to full-body ultrasound scans

The Verge14 h ago
Ultrasound machine screen, close-up
Ultrasound machine screen, close-upPhoto: Pavel Danilyuk / Pexels

Best known as a consumer AI image generator, Midjourney has unveiled its first health product, Midjourney Medical. According to The Verge, the system can perform full-body ultrasound scans with real-time AI guidance and has secured clinical-use clearance from the US Food and Drug Administration.

Midjourney Medical scans tissue with standard ultrasound probes and shows both a traditional image and a deep-learning-guided image. Co-founder David Holz told The Verge, "The same visual network behind our consumer image work can understand ultrasound data." The system runs as a software add-on for existing ultrasound machines.

The company says it used more than 18 million anonymised clinical scans from hospital groups in the US, Europe and Asia to train the model. The team said it paid particular attention to demographic diversity in the training set and balanced sub-specialties such as obstetrics, cardiology and abdominal imaging.

For now the FDA clearance is in the "decision-support" class: a clinician still makes the final diagnosis, but Midjourney Medical can outline organs, flag abnormal findings and suggest probe positioning during a scan to improve image quality. The company said it plans Phase 4 validation trials for autonomous diagnosis within the next 12 months.

Ultrasound is a modality whose image quality is heavily dependent on operator skill. One goal of AI guidance is to narrow this "operator dependency". A 2025 independent assessment by Mayo Clinic found that AI-guided ultrasound raised the scan quality of inexperienced technicians by an average of 31%.

Midjourney's flagship product remains the consumer image-generation tool driven by text prompts. The move into medical imaging is the work of a separate biomedical engineering team built up over the past two years. Holz said, "The consumer AI market is saturating; healthcare is larger by scale and has more unsolved problems."

The system was tested across 14 hospital networks in clinical trials. Published early results show AI-guided scans achieved statistically significant gains in sensitivity for early-stage fatty liver, small thyroid nodules and fetal cardiac anomalies compared with unassisted scans. The false-positive rate remained low.

There is criticism, of course. Dr Curtis Langlotz, a radiologist at Stanford School of Medicine, said: "FDA clearance is important, but AI-guided systems take time to reach their moment in clinical practice." He called for more independent assessment of data set fairness and the cognitive load that flagged findings place on clinicians.

On privacy, Midjourney said the training data was fully anonymised and that no segments allow patient re-identification. Independent auditors have been engaged for compliance with Europe's GDPR and the US HIPAA. The system isolates patient data from Midjourney's consumer AI infrastructure.

Pricing follows a subscription model for healthcare providers, with the entry package at $8,000 per month for twelve devices. Holz said tiered options will be adapted for the health systems of middle-income countries. The company plans to launch pilot programmes in Latin America and Southeast Asia in the coming year.

This article is an AI-curated summary based on The Verge. The illustration is a stock photo by Pavel Danilyuk from Pexels.

Read next