Concurrent Reading

AI-generated regions of interest (ROIs) are superimposed directly onto the mammographic image as the radiologist navigates through the stack. This requires sub-second inference latency to avoid disrupting the visual search pattern. Marks typically appear as bounding boxes, contours, or subtle heatmaps overlaid on suspicious microcalcification clusters, masses, or architectural distortions.

The radiologist retains full control and can toggle AI marks on or off, adjust the sensitivity threshold, or click on a mark to reveal the model's confidence score. This interaction is designed to augment, not replace, the reader. The system functions as a 'second reader' that is consulted simultaneously, allowing the radiologist to accept, reject, or investigate each prompt during the initial interpretation pass.

Unlike a CADe 'second read' paradigm where marks are revealed only after an unaided review, concurrent reading integrates AI into the primary interpretation workflow. This eliminates the need for the radiologist to re-review the entire case. The goal is to reduce inter-reader variability and observational oversights without adding a separate time-consuming step to the diagnostic process.

In Digital Breast Tomosynthesis (DBT) , concurrent reading is particularly critical. As the radiologist scrolls through thin reconstructed slices, AI marks must track the lesion across multiple slices. A Maximum Intensity Projection (MIP) slab may display a synthesized 2D mark, but the system must also indicate the specific slice where the finding is most conspicuous to facilitate efficient correlation.

A core design challenge is preventing inattentional blindness or over-reliance. Marks must be visually salient enough to capture attention for potential misses but not so obtrusive that they distract from unmarked regions or induce satisfaction of search. Effective implementations use subtle, standardized symbology and avoid occluding the underlying tissue morphology.

Concurrent reading systems often link findings across the Craniocaudal (CC) and Mediolateral Oblique (MLO) views. When a mark is displayed on one view, a corresponding geometric indicator may appear on the orthogonal view. This multi-view correlation helps the radiologist rapidly triangulate a true 3D lesion location and dismiss artifacts or overlapping tissue shadows that lack cross-view correspondence.