Astronomy has built pipelines that can sift through vast archives of sky imagery to find moving objects within days of their appearance. Meanwhile, according to Radiology Business, nearly 2,000 CT scans at Walter Reed National Military Medical Center went unread for years. The backlog included approximately 1,300 cone-beam CT scans of patients’ faces and jaws, collected between 2011 and 2016, and was discovered only because a pathologist could not locate a single patient’s file. This is the same archival-image problem, solved in one domain and chronic in another.
The Vera C. Rubin Observatory represents the current state of the art in automated sky surveys. Its pipeline is designed to routinely extract objects from billions of pixels of archival sky imagery. Over its ten-year survey, Rubin is expected to find roughly 10,000 new comets, sifting them from data volumes no human team could plausibly review by eye.
The numbers that sit alongside each other
Ten thousand comets, sifted automatically from a decade of nightly sky imagery. One thousand three hundred facial scans, of living patients, sitting unread on hospital servers for five years.
The pairing is not a budget argument. Space Daily is not suggesting that money allocated to Rubin — a major ground-based observatory funded primarily by the National Science Foundation and the Department of Energy — should have been spent on radiology software. The funding streams are not fungible, the institutional missions are distinct, and the comparison would collapse on contact with any serious treatment of public finance.
What the pairing does surface is an engineering question. The technical problem in both cases is recognizably similar: large volumes of imagery accumulate faster than human reviewers can inspect them; findings of interest — a moving point source against a fixed star field, a lesion against soft tissue — are statistically rare; the cost of missing one is high. Astronomy, working on objects billions of years old and indifferent to delay, has built pipelines that catch them within days. Radiology, working on patients whose findings are time-sensitive, has not built equivalent triage at the same flagship institutions where the problem manifests.
The structural question is why. Ambitious technical programs attract concentrated talent, clear metrics, and protected funding precisely because their goals are bounded and legible. Hospital throughput is none of those things — and the asymmetry recurs across decades and political systems, from Apollo-era complaints about engineering attention directed skyward while urban problems festered, to the present-day gap between observatory pipelines and hospital workflows.
What Rubin actually does, and what radiology actually faces
Honesty about the technical claim matters. Rubin’s pipeline is purpose-built for a specific problem: differencing successive images of the same patch of sky to find what moved or brightened. The objects are points or near-points against a well-characterized background. The false-positive rate is manageable because most candidates can be cross-checked against catalogs of known asteroids and variable stars. The cost of an error is a flagged candidate that turns out to be a cosmic ray hit, which a human can dismiss in seconds.
Medical imaging is harder along several axes that matter for an AI matching a radiologist: CT scans are volumetric, anatomy varies between patients, pathology presents in heterogeneous ways, and FDA clearance imposes validation requirements no astronomical pipeline has to meet. But the Walter Reed failure was not that kind of failure. Thirteen hundred facial scans sat on a server for five years at the country’s flagship military hospital because scans were ordered, acquired, stored, and then nobody was assigned to read them, and no automated system flagged the absence of a read. The triage problem was not to build an AI that can match a radiologist. It was to build a queue that complains when items sit in it for sixty months. That is engineering work substantially less demanding than what Rubin does on a Tuesday.
The asymmetry is not really about technical difficulty. It is about whose problem gets the engineering attention.
The politics of the pipeline
Technical systems embody, in their design choices, decisions about who they serve and who they don’t. The politics of a technology is not only in its use but in its specification: what it is built to do, and for whom.
The Rubin pipeline is built for astronomers, who are organized as a community with shared data standards, shared software repositories, and a cultural expectation that survey data will be processed by common tools. The astronomical community has spent thirty years building toward this — from the Sloan Digital Sky Survey in the 1990s through Pan-STARRS and ZTF to Rubin. The pipeline reflects a community that treated automated archival sifting as a first-class problem and invested accordingly.
Hospital imaging systems are built for billing, scheduling, and the legal record. The DICOM standard that governs medical image storage was designed in the 1980s for interchange between scanner manufacturers and viewing workstations, not for population-scale automated triage. The PACS (picture archiving and communication system) software that hospitals use is sold by vendors whose business model rewards lock-in rather than open processing pipelines. A radiologist who wanted to build a Rubin-equivalent triage system on top of a typical hospital PACS would spend most of the project fighting the data layer.
That is a political fact about the artifact. The pipeline at Rubin and the pipeline at Walter Reed reflect different communities’ decisions about what their image archives are for. Astronomers decided their archives were for discovery. Hospital administrators decided theirs were for compliance. The engineering follows.
What the gap means, and what could close it
It would be easy to read this comparison as an indictment of priorities — to say that a civilization that finds interstellar comets in archival data while leaving facial CT scans unread for five years has its values disordered. Space Daily is not making that argument, because the argument is too tidy. The Rubin pipeline did not crowd out radiology software. The astronomers who built it are not the engineers who would build the hospital equivalent. The funding lines do not connect.
What the comparison does show is that “unexamined imagery” is not a single unsolved problem. It is a solved problem in one domain and a chronic problem in another, and the difference is not technical capacity. The capacity exists. The difference is which communities have organized themselves to demand and fund the pipeline, and which have not.
Rubin will find its 10,000 comets. That is genuinely good. Some fraction of the techniques developed for it will, eventually, migrate into clinical software, as Hubble’s did into mammography. The lag will be measured in decades, and the patients whose scans sit unread during the lag will not benefit from the eventual migration.
Three concrete changes would shorten that lag. First, hospital accreditation bodies — the Joint Commission, CMS — could require what airlines require of maintenance backlogs: an automated queue with aging alarms, where any imaging study unread past a defined threshold escalates without human prompting. The technology to build that has existed for thirty years. Second, the DICOM standard and PACS procurement rules at federal facilities could be revised to mandate open processing interfaces, breaking the vendor lock-in that prevents triage software from being bolted on. Third, the engineering talent that builds survey pipelines — much of it federally funded through NSF and DOE — could be directed, through targeted grant programs, toward porting those pipelines into clinical workflows, rather than waiting decades for organic migration. None of this requires inventing anything. It requires deciding that an unread scan on a hospital server is the same kind of defect as an unprocessed image on an observatory server, and building the institutions that treat it that way.
