Cloud PACS vendors love to talk about elastic storage and anywhere-access. They talk a lot less about what happens when your imaging site has a single, unreliable 25 Mbps connection and a radiologist waiting on a stroke CT. Bandwidth-constrained sites need a different architecture, not just a hope that the internet holds up.
Cache aggressively, at the edge
A local caching gateway that keeps recent and priors studies on-site dramatically cuts what has to move over the wire in real time. Design the cache around your actual read patterns — recent studies and priors for active patients, not a blind LRU policy.
Prioritize by clinical urgency, not arrival order
Study transmission and prefetching should be priority-aware. A STAT trauma series shouldn't be queued behind a routine outpatient study just because it landed in the pipeline five minutes earlier.
Compress deliberately, not by default
Lossy compression can be clinically appropriate for some modalities and unacceptable for others. The right architecture makes compression a deliberate, modality-aware decision — not a blanket setting applied to save bandwidth everywhere.
Plan for the connection going down, not just being slow
Low-bandwidth sites eventually mean no-bandwidth moments. A resilient design queues studies locally and continues acquisition uninterrupted, syncing to the cloud PACS once connectivity returns — rather than blocking the modality on a live connection.
Monitor the actual site, not just the cloud dashboard
Cloud PACS vendor dashboards show cloud-side health. They don't show you that a specific site's local gateway has been silently retrying for two hours. Independent monitoring at the edge catches problems the vendor's dashboard never will.