Radiology AI drives $600K-plus offers

Radiology AI is having a very specific moment. Money is flowing in, hospitals are chasing productivity, and vendors are trying to move from single-algorithm demos to full workflow platforms. But the gap between the sales pitch and the real hospital environment is still the whole story. The clearest sign of that came on March 2, when RadNet said it would buy Gleamer for up to $269.3 million and fold it into DeepHealth — basically betting that imaging AI is now valuable enough to justify platform-scale deals. (radiologybusiness.com) ### Why is radiology such a hot AI market? Radiology is almost built for software. The inputs are already digital, the workloads are huge, and the staffing squeeze is real. That is why radiology now makes up roughly 80% of FDA-cleared medical AI algorithms in the U.S. But clearance is not the same thing as routine use — or payment. (radiologybusiness.com) ### What changed this year? The market stopped acting like a collection of clever point tools and started acting like an enterprise software business. RadNet’s Gleamer deal is the cleanest example. RadNet said Gleamer should generate about $30 million in annual recurring revenue in 2026, and the combine(radiologybusiness.com)rkflows, and recurring revenue — not just model accuracy screenshots. (radiologybusiness.com) ### So are hospitals actually getting value? Sometimes, yes. But the value usually comes from narrow, boring workflow gains first. Northwestern Medicine’s in-house generative AI system cut radiograph documentation time by 15.5% across 11,980 model-assisted interpretations in live care, with no meaningful drop in radiolog(radiologybusiness.com), but by shaving friction out of repetitive work. (jamanetwork.com) ### Why doesn’t that automatically turn into big revenue? Because hospitals do not get paid just because an algorithm exists. Even though radiology dominates FDA-cleared AI, only two Category 1 CPT payment codes for newer AI existed as of January 2026, and both were for cardiac imaging applications. Most vendors still have regulatory clearance without reimbursement(jamanetwork.com) not a clean line item that says “AI reimbursement.” (radiologybusiness.com) ### Where does the hype break? At the messy edge cases. Real clinical environments are full of low-quality images, atypical anatomy, rushed acquisition, legacy PACS, and workflows nobody designed from scratch. A 2026 real-world study of IDx-DR — one of the best-known autonomous imaging AI systems — show(radiologybusiness.com)able. The system still performed well on severe disease, but only after image quality cooperated. That is the catch in one study. (nature.com) ### Why are vendors shifting to workflow? Because standalone algorithms are harder to justify now. Signify Research’s read on the RSNA 2025 market was that vendors are moving toward orchestration, analytics, and tighter PACS integration. Basically, if an AI tool does not fit the existing imaging stack, it becomes one more screen, one more alert, and one more procurement headache. (radio([nature.com)intelligence/radiology-ai-vendors-shift-focus-workflow-integration-and-enterprise-value)) ### Does this mean radiologists are getting replaced? Not in the broad way people keep implying. The current winners are mostly assistive tools, triage systems, and narrow autonomous products. The practical market is about force multipliers — faster reporting, better prioritization, fewer misses, smoother follow-up. Full replacement still runs into the same wall: medicine is messy, and the long tail of weird cases is where confidence breaks first. (nature.com) ### Bottom line? Radiology AI is no longer just a science project. The deal sizes are real, the productivity gains can be real, and the market is consolidating fast. But the durable value is showing up in integrated workflow software, not magic autonomy — at least not yet. (radiologybusiness.com)