A coordinator oversees three staff members per modality, who cite the literature and allow the clinician to make corrections: an AUROC of 0.80 for structural progression and 0.68 for pain.
Predicting which patients will see their knee osteoarthritis get worse remains difficult, and existing models often deliver a score without justification. Researchers propose OA-MAP, a multi-agent system in which an MRI agent, an X-ray agent, and a clinical officer each analyze their data, select their own tools, and search for supporting publications under the supervision of a coordinating officer.
A human in the loop when the doubt is strong
The device incorporates a mechanism based on uncertainty: when conclusions are fragile, the clinician is asked to validate or correct intermediate observations, which triggers a new calculation of risk. The final result therefore links the estimate to legible findings, crossed between modalities and backed by references.
Evaluated from 600 participants in the FNIH consortium cohort on osteoarthritis biomarkers, of which 100 were reserved for testing, OA-MAP reached an AUROC of 0.80 for structural progression and 0.68 for pain progression, a symptom known to be more difficult to anticipate from imaging.
Read the full article on the original source
Go to the source →