Intracranial hemorrhage annotation on head CT
Bounding-box annotation and characterization of intracranial hemorrhages on non-contrast head CT, with AI-assisted review.
An AI vendor building a head CT triage product needed radiologists to annotate intracranial hemorrhages on non-contrast head CT acquisitions, handling continuous versus discontinuous bleeds, multiple compartments and AI-prediction adjudication.
Annotate every visible intracranial bleed with a reconstructed 3D bounding volume, characterize each observation (type, location, density), and adjudicate the platform's AI predictions using a defined confirm / uncertain / reject workflow.
Radiologists were trained on the client's browser-based tool, including the specific rules for continuity, multi-compartment bleeds, exam and series characterization, and shortcut-driven adjudication of AI predictions. Difficult cases were flagged with structured comments for the client's review.
The client received consistent, well-characterized annotations with clear handling of edge cases such as artifacts, post-op studies and multi-compartment bleeds. The adjudication workflow accelerated iteration on their detection model.