AI Takes Over the Laborious Tracking of Oncology Imaging, With Moffitt Testing the Waters in Research Settings First

The challenge of interpreting cancer imaging often involves more than finding a single tumor. It requires identifying multiple lesions in whole-body scans, then comparing each one with images taken months earlier to determine whether they have shrunk, grown, or newly appeared. French medical AI company Raidium has now brought these steps into a single workflow and deployed it at Moffitt Cancer Center in the United States, initially to support oncology research and clinical trials.

The system, called Raidium Read, combines whole-body lesion detection, AI image segmentation, and longitudinal tracking. When reading new images, the system can transfer lesions marked in previous examinations to the current images and automatically generate RECIST measurements. RECIST is a tumor response assessment standard commonly used in clinical trials that primarily evaluates treatment efficacy based on changes in lesion size.

Moffitt selected the system to replace its existing radiomics tools. For research teams, its potential value lies in reducing the manual work involved in repeatedly locating, outlining, and measuring lesions, while making it easier to connect records across different time points. Raidium says this standalone research workflow does not require hospitals to first complete back-end system integration, helping lower the barrier to adoption in trial settings.

However, “deployed at a cancer center” does not mean the system is available for routine clinical diagnosis. Access currently remains limited to clinical trials and oncology research. Physicians cannot treat the relevant features as clinical tools approved by regulators solely on the basis of this deployment. Raidium is seeking US FDA 510(k) clearance for some of the features and expects to announce progress by the end of 2026; the final timeline will still depend on the review outcome.

Performance evidence must also be considered separately from the product narrative. The company claims that automated RECIST measurements can reduce variability among readers threefold, but neither the announcement nor two reports covering the same event disclosed the study sample size, cancer-type composition, imaging sources, comparison benchmark, or external validation results. Without this information, it is not currently possible to determine whether the benefits can extend across different scanning equipment, medical institutions, and complex lesion types.

This deployment therefore more closely resembles a research test conducted in a real-world work environment: Rather than making only a single-point assessment, the AI attempts to connect lesion detection, contouring, and treatment-response tracking. Whether it can genuinely save physicians time, maintain measurement consistency, and safely extend into routine care must still be answered through independent validation, real-world usage data, and FDA review.