Raidium brings AI oncology imaging tool to US cancer centers

Reading scans in oncology is grinding work. Radiologists tracking tumors across dozens of follow-up studies have to manually hunt for lesions, measure them, compare them to prior images, and document everything, often inside software that has barely changed in two decades. It is slow, repetitive, and error-prone in ways that matter for patients.

Paris-based startup Raidium thinks it has a better way. The company has launched its AI imaging platform, called Raidium Read (R.Read), in the US, starting with oncology research centers. Its first confirmed deployment is at Moffitt Cancer Center in Tampa, Florida, one of the country's top-ranked cancer research institutions, where it was brought in to replace an existing radiomics tool.

R.Read is not a plugin bolted onto an existing system. Raidium built its viewer from scratch with AI at the core, targeting the specific pain points of oncology imaging: finding lesions across the whole body, tracking them over time, and measuring them consistently without forcing radiologists to switch between tools.

How does it work?

The platform runs what Raidium calls an agentic AI approach, meaning the system can carry out multi-step imaging tasks with minimal manual input. In practice, that means radiologists get:

• Whole-body lesion detection across organ types
• Automated segmentation of lesions using foundation models
• Longitudinal tracking that transfers lesion data across follow-up and prior studies automatically
• Automated RECIST measurements, the standard used in oncology trials to assess whether tumors are growing or shrinking, with the company claiming a threefold reduction in variability between readers

The goal is to reduce the manual work involved in oncology imaging review while keeping the radiologist in control of every clinical decision. Raidium says cancer centers can get started without any integration prerequisites, which lowers the barrier to adoption for centers that do not want to overhaul their existing IT infrastructure.

R.Read is available now for clinical trial and oncology research use. Raidium is pursuing 510(k) clearance with the FDA for a subset of features and expects to announce that clearance before the end of 2026. Full use in routine clinical practice is pending that regulatory approval.

Why does it matter?

Radiology is the single biggest category for FDA-approved AI medical devices, accounting for more than 70% of all cleared tools according to industry estimates. But having lots of approved AI products has not translated into smooth adoption. A 2025 Philips Future Health Index survey found that 41% of radiologists feel current AI tools do not actually address their real-world needs.

The core complaint is fragmentation. Most AI tools are narrow solutions that add steps rather than remove them, requiring radiologists to jump between systems and rebuild context each time. Raidium's pitch is that building the viewer and the AI together from the start produces something more useful than adding AI features to old software.

Oncology is a smart place to prove that. Tumor tracking is one of the most time-consuming parts of a radiologist's workload, and the stakes are high. Research suggests that earlier and more precise lesion detection can reduce the need for surgery in a significant proportion of patients. The WHO is also updating its guidelines around screening and detection of precancerous lesions, which will put more pressure on imaging workflows globally.

Cesar Lam, a radiologist at Moffitt's Diagnostic Imaging and Interventional Radiology Department, said the platform has opened up research projects his team could not have attempted before: "It is transforming and empowering how we conduct oncology clinical research projects, giving our teams a tool designed for the complexity of real-world imaging data."

The context

Raidium was founded in Paris and has offices in Silicon Valley. The company combines a clinical viewer product with an internal AI research lab that developed a foundation model called Curia, which it says achieves strong performance across imaging tasks. A peer-reviewed paper on Curia is forthcoming in the journal Radiology AI.

The US launch puts Raidium up against a crowded but still unsettled market. Large incumbent vendors like Philips, GE HealthCare, and Siemens Healthineers are all pushing AI into their imaging platforms, while a wave of startups including Viz.ai, Annalise.ai, and Intelerad are targeting specific workflow problems in radiology. What most of them share is the same architectural challenge Raidium is trying to sidestep: they built AI on top of existing systems rather than redesigning the system around AI.

Cancer centers interested in R.Read can register on an invitation basis at raidium.eu/viewer.