The platform delivered diagnostic imaging to radiologists at home. It ran the whole of this tenure and was deployed across several hospitals in North Carolina.
That is telemedicine, fifteen years before the market used the word for it.
The economics that drove it
A radiologist is expensive and scarce. Requiring one to be physically present to read a study means either carrying more of them on staff than the caseload justifies, or making the read wait.
Delivering the study to the radiologist’s home instead let a hospital hold a smaller staff and still reach a reader, and reach third-party radiologists elsewhere in the state. It also gave a referring physician a route to a second opinion that did not depend on who happened to be in the building that day.
The requirement that set everything else
The radiologist at home had to have the same tools they would have had at the hospital.
A diagnostic read is not a preview. If the reading environment is degraded the read is worthless — and worse than worthless if somebody acts on it. Every other decision in the design is downstream of that sentence.
Whatever the transport at that address could provide
This was the late 1990s, so the link was ISDN, dial-up or frame relay, chosen per physician by what was actually available at their address rather than by what the design would have preferred. You do not get to specify the last mile into a radiologist’s house.
That made compression an architectural decision rather than a bandwidth optimization, because lossy compression on a primary diagnostic read is a clinical and regulatory question before it is an engineering one. And studies were pushed or pre-fetched ahead of the read, so the transfer time sat outside the radiologist’s working session instead of inside it — the oldest trick for a slow link, and still the right one.
The hard part was the hospital end
Not the link. Tying the platform into existing imaging estates, and particularly into the write-once optical archives, where a request for a prior study meant a physical fetch from a jukebox. Retrieval latency of that kind has to be designed around rather than wished away: a robot arm is not going to get faster because the schedule needs it to.
A regulated device, not software
The platform operated under Food and Drug Administration clearance. My contribution was the data gathering and requirements work feeding that process rather than authoring the regulatory filing itself.
What matters for an engineer is the constraint it imposes. The design is part of a regulated device, so it cannot be revised casually — a change that would be a ticket in enterprise software is a controlled event with evidence attached. That is a different engineering culture from the one most infrastructure work happens in, and learning it at the start of a career sets a default that never entirely goes away.
Outcome
A hospital could hold a smaller radiology staff and still reach a reader, and a referring physician had a route to a second opinion that did not depend on who was in the building.