Fracture characterization on limb X-ray
Bounding-box annotation of fractures on multi-view X-ray series with type and healing status attributes, on a cloud annotation platform.
An AI vendor building a fracture triage model needed radiologists to annotate fractures on X-ray series covering multiple incidences of the same anatomy (e.g., face and profile views of a hand), with consistent characterization across all views of a series.
Draw tight bounding boxes around each fracture, correct pre-annotated fracture type (displaced vs. non-displaced) and add a healing status attribute (acute, partially healed, healed), keeping attributes consistent across incidences of the same series.
A dedicated pool of radiologists was onboarded on the client's browser-based tool and trained on the fracture healing continuum (hematoma → granulation → soft callus → hard callus → compact bone) with reference imagery. Series grouping was used to characterize the same fracture across views, and a 'partially healed' fallback was applied for ambiguous cases.
The client received consistently characterized fractures across multi-view series, with tight bounding boxes and reliable status attributes. Pre-annotation errors on 'type' were systematically corrected before delivery.