Search papers, labs, and topics across Lattice.
MIRROR addresses the dual challenges faced by radiologists when interpreting model outputs by implementing a structured approach that combines a multi-label classifier, a Grad-CAM localizer, and a report writer that operates independently of the image data. This design ensures that the generated reports are strictly based on the classifier's outputs, preventing the introduction of unverified claims. The system achieves a macro AUROC of 0.729 on ChestMNIST, demonstrating improved precision over random ranking, though it reveals significant limitations in positive predictions under typical class imbalances in radiology.
MIRROR exposes the critical flaws in radiology AI reporting by ensuring that generated text can only reflect what the model has actually classified, not what it might imply.
A radiologist reading a model's output faces two problems. The model returns a number and no reason, and any system that turns that number into readable prose can quietly add claims the model never made. MIRROR is a research prototype built to separate those failures. It chains a multi-label classifier, a Grad-CAM localizer that turns each positive finding into a named anatomical region, and a report writer that receives the labels, probabilities, and regions but never the image. Because the language layer cannot see pixels, it cannot assert a finding the classifier did not make. We are precise about what that buys: a MIRROR report's findings are auditable against the probability vector, while the sentences framing them are ordinary generated text, and we show one stating a cardiothoracic ratio the system never measured. One registry holds the taxonomy, anatomy, and phrasing for chest X-ray, brain MRI, and head CT, so adding a modality is a data change; all three are routed and tested, one is trained. On ChestMNIST that classifier reaches macro AUROC 0.729 and ranks better than chance on all 14 labels, at 1.6 to 6.8 times the precision a random ranker would get. Yet at the default 0.5 threshold it emits no positive prediction at all for 11 of them, and its excellent-looking Brier score of 0.045 sits beside the 0.047 earned by a predictor that ignores the image. The discrimination is real; the decisions are not. Under the class imbalance normal in radiology, aggregate metrics flatter models that do nothing, and should be reported against that floor.
