Raidium Read: AI-native radiology viewer debuts at Moffitt, signaling a PACS-free workflow path for RECIST tracking

Raidium, a Paris and Silicon Valley-based startup, has launched its AI-native radiology platform Raidium Read at Moffitt Cancer Center. Instead of bolting AI onto a legacy PACS viewer, the company built the viewer from scratch around its Curia foundation model, enabling automated RECIST tumor measurements across organs and time points without backend PACS integration. For AI builders in healthcare, this model-centric approach signals a faster path to clinical deployment and a potential reduction in inter-reader variability.

What happened

Raidium Read replaced Moffitt's legacy radiomics applications and is currently available for clinical trials and research use in the US. FDA 510(k) clearance is expected before the end of 2026. The system is built around Curia, Raidium's proprietary foundation model trained on over 200 million CT and MRI slices from 150,000 exams. Curia performs organ-agnostic, automated RECIST measurements across multiple time points. Raidium says this cuts inter-reader variability by a factor of three. "For twenty years, the standard PACS viewers have resisted evolution," said Paul Herent, Raidium's CEO and co-founder. Dr. Cesar Lam, a radiologist at Moffitt, said the platform enables research projects that "would have seemed impossible not too long ago."

Why AI builders should care

The key architectural decision is embedding the model inside a purpose-built viewer rather than layering AI on top of an existing PACS. This reduces deployment friction because no backend integration is required. For teams building AI tools for clinical workflows, this pattern could accelerate adoption. The system automates lesion detection, segmentation, and historical mapping across follow-up scans, addressing a tedious manual task that is time-consuming and inconsistent between readers.

Practical implications

Raidium Read automates RECIST measurements across multiple time points, reducing inter-reader variability by about threefold. The system scans large-volume imaging inputs, detects and segments lesions across anatomical regions, and maps historical lesion data against new follow-up scans. Deployment is faster than traditional PACS installations because no backend integration is needed. This could enable research projects that were previously impractical, as noted by Dr. Lam.

Caveats

The evidence for these capabilities comes from company materials and press coverage, not independent third-party validation. FDA 510(k) clearance is anticipated but not yet granted; the product is currently for clinical trials and research use only. Details about product capabilities are drawn from the parent article and excerpts rather than independently verified benchmarks. Builders should treat performance claims as vendor-reported until peer-reviewed studies or regulatory filings confirm them.