RESEARCH / MEDICAL AI
Describes how language generation, image localisation and explicit rules can be combined into traceable draft mammography findings. It examines disagreements between models, auditability and the evaluation still needed before clinical integration.
Read on this pageA model that writes a plausible mammography finding and a model that locates an image feature have different strengths. This HOPPR-authored paper examines an orchestration layer that reconciles their outputs into draft findings under explicit rules, while keeping each decision traceable.
The design combines generated clinical language with image-grounded geometry and uses code to constrain assessment changes. The paper is unusually explicit about unresolved issues, including how uncorroborated detections may affect recall and how input-contract mismatches can affect measurements. It argues for evaluation by the type of reconciliation action, rather than a single aggregate score. For readers assessing multimodal medical AI, the contribution is the decision framework and the evaluation still needed to establish its value.
Generated narrative and image localisation are reconciled according to explicit rules.
Each reconciliation is assigned an action code that can be audited and evaluated.
The paper documents unresolved policy and measurement-contract issues alongside the design.
Authored by HOPPR. Results against the pipeline's stated performance targets are not yet documented and are not claimed. Natural-prevalence evaluation and resolution of the identified confidence, policy and measurement issues remain necessary before clinical integration.
The complete paper, with its original figures, tables and references.
31 pages · Read without leaving this pageDiscuss the approach, evidence and relevance to your team with the people building it.
