The AI-Native Revolution in Oncology: Raidium’s Disruptive Entry into US Radiology

The landscape of medical imaging is currently undergoing its most significant shift since the transition from physical film to digital Picture Archiving and Communication Systems (PACS) in the late 1990s. At the center of this transformation is Raidium, a Paris and Silicon Valley-based startup that recently announced the US launch of its AI-native radiology platform at the prestigious Moffitt Cancer Center. By moving away from the traditional "bolt-on" approach to artificial intelligence, Raidium is attempting to redefine the foundational workflow of the radiologist, promising to solve the chronic issues of inter-reader variability and the sheer manual labor involved in tracking tumor progression.

Main Facts: A New Paradigm for Cancer Imaging

Raidium has officially deployed its flagship platform, Raidium Read, at Moffitt Cancer Center, one of the premier oncology research and treatment institutions in the United States. Unlike previous iterations of artificial intelligence in healthcare, which typically exist as secondary applications or "widgets" within a legacy PACS environment, Raidium Read is an "AI-native" viewer. This means the artificial intelligence is not an external tool; it is the core engine upon which the viewing interface is built.

The platform is powered by Curia, a proprietary foundation model developed by Raidium. Curia was trained on an immense dataset comprising over 200 million CT and MRI slices sourced from 150,000 comprehensive exams. This scale allows the model to be "organ-agnostic," meaning it can identify, segment, and measure lesions and anatomical structures across the entire body without requiring specific, narrow algorithms for every different organ.

The primary function of the platform at Moffitt is to automate the tracking of tumor response to treatment using the RECIST (Response Evaluation Criteria in Solid Tumors) standard. In clinical trials and routine oncology, RECIST is the gold standard for determining if a patient’s cancer is progressing, stable, or in remission. However, performing these measurements manually is notoriously tedious and prone to human error. Raidium’s platform automates this process, reportedly reducing inter-reader variability—the discrepancy between how two different doctors might measure the same tumor—by a factor of three.

Currently, Raidium Read is being utilized at Moffitt for clinical trials and research purposes. While it is already making an impact in the laboratory and trial setting, the company is moving toward full clinical implementation, with FDA 510(k) clearance anticipated before the end of 2026.

Chronology: From Legacy PACS to Foundation Models

To understand the significance of Raidium’s launch, one must look at the two-decade stagnation of radiology software. Since the early 2000s, the "legacy PACS" has been the industry standard. While these systems successfully digitized the darkroom, they have remained largely unchanged for twenty years, acting essentially as digital lightboxes with basic measurement tools.

2015–2020: The Era of Narrow AI
During this period, the medical community saw a surge in "narrow AI" tools—algorithms designed to do one specific thing, such as detect a lung nodule or identify a brain bleed. While useful, these tools created "app fatigue." Radiologists had to leave their primary workflow, open a separate window, and manually transfer data back into their reports. This friction prevented widespread adoption.

2021–2023: The Development of Curia
Recognizing that the bottleneck was the interface itself, Raidium began developing Curia. Instead of training a model to find one specific disease, they aimed to create a "foundation model" for radiology. By training on 200 million slices, they taught the AI the fundamental "language" of human anatomy in three dimensions. This period involved rigorous validation to ensure the model could handle the nuances of different scanner manufacturers (GE, Siemens, Philips) and varying image qualities.

2024: The Moffitt Partnership and US Expansion
The launch at Moffitt Cancer Center represents Raidium’s strategic entry into the American market. Moffitt, known for its high-volume clinical trials, provided the perfect environment to test if an AI-native viewer could actually speed up the "longitudinal tracking" (comparing scans over months or years) that is the backbone of oncology.

2025–2026: The Path to FDA Clearance
The current phase involves gathering real-world evidence from the Moffitt deployment to support its FDA 510(k) submission. This timeline reflects a cautious but steady march toward making AI-native viewing the standard of care for clinical diagnosis across the United States.

Supporting Data: Quantifying the Impact on Radiology

The "3x reduction in inter-reader variability" cited by Raidium is perhaps the most critical metric for the future of oncology. In clinical trials, a difference of just a few millimeters in a tumor measurement can be the difference between a drug being labeled a "success" or a "failure."

The Problem of Human Variability

Studies have shown that when two different radiologists look at the same CT scan of a lung tumor, their measurements can vary by as much as 15-20%. This variability stems from:

• Slice Selection: Choosing a slightly different "slice" of the 3D scan to measure.
• Boundary Definition: Where exactly the tumor ends and healthy tissue begins.
• Fatigue: Radiologists may review up to 100 cases a day, leading to "satisfaction of search" errors.

The Curia Solution

Raidium’s Curia model addresses this by automating the segmentation. Because the AI views the entire volume of the scan simultaneously, it identifies the largest diameter of a lesion with mathematical precision that remains constant regardless of how many cases the system has processed.

Deployment Efficiency

Traditional PACS installations are notorious for requiring massive backend infrastructure, server rooms, and months of IT integration. Raidium’s platform is designed to be "backend agnostic." By requiring no deep backend integration for its initial research deployments, the system can be "turned on" in a fraction of the time it takes to install traditional medical software. This agility is a key differentiator in a healthcare market that is increasingly wary of long, expensive IT projects.

Official Responses: Voices from the Frontline

The leadership at both Raidium and Moffitt Cancer Center have framed this launch as a philosophical shift in how technology serves medicine.

Paul Herent, CEO and Co-founder of Raidium, has been vocal about the failures of the previous generation of software. “For twenty years, the standard PACS viewers have resisted evolution,” Herent stated. He argues that by "bolting" AI onto old systems, the industry has ignored the user experience of the radiologist. Herent’s vision is to make the AI invisible—so deeply integrated into the viewer that the radiologist doesn’t feel like they are "using AI," but rather using a "smarter" version of their traditional tools.

Dr. Cesar Lam, a radiologist at Moffitt Cancer Center, highlighted the practical implications for high-level research. He noted that the platform enables research projects that “would have seemed impossible not too long ago.” In a research context, the ability to retroactively analyze thousands of past scans with automated, consistent measurements allows for "big data" oncology research that would have previously required thousands of hours of manual labor by highly paid specialists.

Industry analysts have compared Raidium’s approach to other "reasoning-based" AI transitions. Much like how Corti’s Symphony AI revolutionized medical coding by treating it as a reasoning problem rather than a simple labeling task, Raidium is treating radiology as a holistic data-mapping problem.

Implications: The Future of the "Radiology Cockpit"

The success of Raidium Read at Moffitt has profound implications for the broader healthcare ecosystem, stretching from clinical trial efficiency to the daily workload of hospital staff.

1. Accelerating Drug Discovery

In the world of pharmaceutical development, time is the most expensive variable. By automating RECIST measurements, Raidium can potentially shave weeks or months off the data-collection phase of clinical trials. If the measurements are more accurate and less variable, the "statistical noise" in a trial is reduced, potentially allowing for smaller, more efficient studies.

2. Combating Radiologist Burnout

The global shortage of radiologists is a looming crisis. By automating the "tedious" parts of the job—finding the same lesion on three different scans from three different years and measuring them—Raidium allows radiologists to focus on high-level diagnostic reasoning. This shift from "measurer" to "consultant" is seen as a vital step in making the profession sustainable.

3. The Move Toward "Foundation Models" in Medicine

Raidium’s use of a foundation model (Curia) marks a departure from the "one-task-one-algorithm" era. This suggests a future where a single medical AI can assist in a wide variety of tasks—from detecting a fracture to grading a rare sarcoma—within a single, unified interface.

4. Regulatory and Ethical Considerations

As Raidium moves toward its 2026 FDA clearance goal, the focus will shift to "explainability." While the AI provides a measurement, the radiologist must still be able to see why the AI made that choice. Raidium’s native viewer approach facilitates this by allowing the doctor to interact with the AI’s segmentations in real-time, maintaining the "human-in-the-loop" necessity of modern medicine.

In conclusion, the launch of Raidium Read at Moffitt Cancer Center is more than just a software update; it is a challenge to the status quo of medical imaging. By proving that an AI-native architecture can reduce variability and simplify complex workflows, Raidium is setting the stage for a new era where the "digital lightbox" finally becomes as intelligent as the doctors who use it.