Across the network's 16 centres, 3D mammography had become the standard screening exam, and each one produces hundreds of image slices for a radiologist to review, far more than an older 2D study. Reading times ran well above what 2D screening required, and a radiologist shortage meant fewer people were available to absorb that extra volume.

Every exam was read strictly in the order it arrived, with no way to tell a likely-abnormal case from a routine one before a radiologist opened it. That queue discipline was fair in principle, but it meant a genuinely concerning case could sit behind routine exams for exactly as long as everything else.

The unread backlog pushed past three weeks, and every exam waiting in it delayed two things at once: the results letter a patient was waiting for, and any follow-up scheduling that depended on that letter going out. The longer the queue grew, the further behind those two things fell.

We integrated a regulator-cleared AI detection engine directly into the network's existing reading workflow rather than adding a separate system. Every 3D exam now receives a case-level suspicion score as it comes in, and that score feeds a prioritised worklist so likely-abnormal exams surface ahead of routine ones instead of waiting their turn in arrival order.

Finding overlays render inside the same viewer radiologists already use, so the AI output sits alongside the images rather than in a separate application. Every AI-assisted case still requires a radiologist's sign-off before a result goes out, keeping the reading decision with the radiologist and the software in a supporting role.

A network-wide dashboard tracks turnaround, recall rates and reading volumes across all 16 centres, giving the network one view of how the backlog is moving and whether prioritisation is working the way it was meant to.