RBfracture reduces the noise by flagging only the fractures that need attention

Radiobotics built its RBfracture AI tool around the assumption that the most valuable thing an AI tool can do is tell the clinician what actually matters, right now. Clinical product lead Rikke Bachmann explained: “We made a design decision at the heart of RBfracture that we are asked about more than almost any other feature: why doesn’t RBfracture detect healed fractures?

“When a radiologist or clinician sees an examination, they are working in a context. In acute care settings, the question is almost never has this patient ever broken something? Instead the question is whether there is an injury that needs attention right now. These are fundamentally different clinical questions, and they deserve fundamentally different tools.”

Bachmann pointed out that the problem with flagging healed fractures is that it does not just add information; it adds noise. “When a chronic finding gets surfaced alongside acute ones, it creates ambiguity when clarity matters most. Does this patient need treatment? Was this finding missed? Should we act?
“In busy clinical workflows, this kind of uncertainty has real consequences for the clinician’s time, for the patient’s care pathway and, ultimately, for the system’s resources.”

Bachmann continued: “We’ve observed consistently across fracture detection tools in the market that the default approach is to flag everything. The assumption seems to be that more findings means more safety. But over-calling is not neutral. It erodes trust in AI-assisted workflows and, in the worst cases, leads to unnecessary intervention. Designing for specificity over sensitivity in this context wasn’t the easiest path. But it was the right clinical call, and the call we’ll make every time.”

Read this report on page 4 of the September 2026 issue of RAD Magazine.