Imaging Informatics

Imaging informatics is the branch of medical imaging concerned with the information around the picture: how images are acquired, moved, stored, found, displayed, described, and acted upon inside a hospital or a health system.

12 frameworks, from 1970 to today · Medical Imaging · Medicine & Health

still a working view ended · each leads to its entry below

Overview

Imaging informatics is the branch of medical imaging concerned with the information around the picture: how images are acquired, moved, stored, found, displayed, described, and acted upon inside a hospital or a health system1. A radiologist looking at a chest CT sees the anatomy; the imaging informaticist sees a study that must be ordered, scheduled, matched to the right patient, transmitted over a network, archived for years, retrieved in seconds, read against prior exams, and turned into a report that a referring physician can act on. The field's central problem is that a modern hospital generates enormous numbers of images—a single CT can produce thousands of slices—and the clinical value of those images depends entirely on the machinery that delivers them to the right person at the right time. Imaging informatics is the discipline that builds and studies that machinery.

Its history is largely a history of standards, systems, and the slow conversion of radiology from a film-based craft into a data-intensive enterprise. The turning points below follow that conversion.

Film, folders, and the first tracking systems

For most of the twentieth century, a radiology department ran on film. Exposed films were developed, hung on light boxes, filed in envelopes, and retrieved by hand from a film library. As imaging volume grew after the 1960s, the film library became a bottleneck: a lost film meant a repeated exam or a delayed diagnosis, and a large department might hold hundreds of thousands of folders. The first informatics response was the radiology information system (RIS), a hospital software system that managed patient registration, exam ordering, scheduling, and reporting for the imaging department. Early RIS installations appeared in the 1970s and 1980s, often as modules of a hospital information system, and they introduced the idea that an imaging exam is a record with a status—ordered, scheduled, performed, reported—rather than a physical object.

Film tracking followed: barcode labels and later radio-frequency identification tags let staff record where each film jacket was. These systems were modest, but they established the field's foundational belief that the image is only useful if its metadata—who, what, when, where—travels with it reliably.

Digital acquisition and the PACS problem

The decisive shift came with digital modalities. Computed tomography, introduced clinically in the early 1970s, produced numerical data rather than film, and by the 1980s ultrasound, nuclear medicine, and eventually MRI were also digital. Digital data could in principle be stored and transmitted electronically, but early systems were vendor-specific islands: a scanner's console could display its own images, and nothing else could read them.

The answer was the picture archiving and communication system (PACS): a system that stores images digitally and delivers them over a network to viewing workstations2. Early PACS research and prototypes appeared in the 1980s, with notable academic and military-funded efforts in the United States and Europe, and clinical PACS spread through the 1990s as storage and network costs fell. A PACS only works if every device speaks the same language, and two standards made that possible. DICOM (Digital Imaging and Communications in Medicine), first published in 1985 and maintained since by a standing committee, defines how images and their metadata are encoded and exchanged; its header carries patient identity, study details, and acquisition parameters alongside the pixel data. HL7 (Health Level Seven), whose first version appeared in 1987, standardizes the messages that carry orders, results, and patient demographics between hospital systems. The pairing—HL7 for the textual workflow, DICOM for the images—became the backbone of radiology informatics.

The practical consequence was that a radiologist could sit at a workstation and call up a patient's prior study from another building, compare it side by side with today's, and dictate a report without touching film. This changed reading rooms, staffing, and the geography of radiology: subspecialty interpretation could be centralized, and later outsourced across time zones.

Workflow and the human at the workstation

Once images arrived on screens, the question became how to make the reading process efficient and safe. Workflow and human-computer interaction research in imaging informatics studied reading-room ergonomics, the design of hanging protocols (the rules that decide how a study's images are laid out on screen), worklist prioritization, and the cognitive load of reading hundreds of images per shift. A recurring finding was that much of a radiologist's time went not to interpretation but to navigation—finding the right study, the right prior, the right series. Speech recognition, introduced into radiology reporting in the late 1990s and 2000s, replaced much transcription and changed report turnaround, though it also introduced new editing burdens. These studies treated the radiology workstation as a designed artifact, and their methods—time-motion analysis, eye tracking, usability testing—drew on human factors engineering as much as on medicine.

Computer-aided detection and diagnosis

A parallel line of work asked whether computers could help find the abnormality, not just display it. Computer-aided detection (CADe) and computer-aided diagnosis (CADx) use algorithms to mark suspicious regions or estimate the likelihood of disease3. The best-known early success was in mammography: CAD systems for breast cancer screening were developed from the 1980s, and a landmark clinical study by Chan and colleagues in the late 1980s showed that a computer could detect microcalcifications and masses on digitized mammograms at a level comparable to radiologists. The U.S. Food and Drug Administration approved the first commercial mammography CAD system in 1998, and CAD was soon adopted widely in screening.

The results were mixed. CAD increased the number of findings flagged, but studies through the 2000s questioned whether it improved cancer detection or reduced recalls, and some suggested it increased false positives and biopsy rates. The episode taught the field a lasting lesson: a detection algorithm is only as good as the workflow that consumes its output, and a flagged region that a radiologist must dismiss is a cost, not a benefit. CAD's methods—handcrafted features, region growing, classifier ensembles—were the direct ancestors of the deep-learning systems that followed.

Structured reporting and the report as data

Radiology's product is the report, and for decades that report was free text. Structured reporting is the movement to make reports machine-readable: standardized templates with defined fields, controlled vocabulary, and explicit findings, so that a report can be searched, aggregated, and compared. The idea has roots in the 1990s and gained institutional force through the Radiological Society of North America's reporting initiative in the 2000s, which produced standardized templates for common exams. The deeper goal is the report as data: if a finding like "nodule, 8 mm, upper lobe" is a coded field rather than a sentence, it can feed a registry, trigger a follow-up recommendation, or train a model. Structured reporting has spread unevenly—templates are widely used for screening and cancer staging, less so for general diagnostic work—but it reframed the report as an information object rather than a document.

Enterprise imaging and the cloud

As digital imaging matured, the problem shifted from a single department to the whole enterprise. Images were no longer only radiology's: cardiology, pathology, dermatology, and ophthalmology all generated images, often in separate systems with separate archives. Enterprise imaging is the effort to unify these into one governed archive with a shared viewer and shared patient identity, so that a clinician can see all of a patient's images in one place. The cloud accelerated this: instead of buying and maintaining on-site archives, hospitals could store images in vendor data centers, with the archive scaling elastically and accessible to remote readers. Cloud PACS and vendor-neutral archives (systems that hold images independently of any one viewing application) became common in the 2010s and 2020s, and the COVID-19 pandemic sharply increased demand for remote reading and image sharing between institutions.

The learning health system and imaging data science

The accumulation of digital images, reports, and metadata created a new possibility: the imaging enterprise as a source of evidence. The learning health system idea—that routine care should continuously generate data that improves care—was articulated in the 2000s and applied to imaging through registries, quality metrics, and large annotated datasets. Imaging data science grew around these resources: the public release of large image collections, the rise of radiomics (extracting quantitative features from images), and the use of natural language processing on reports to build labels at scale. The field's practitioners began to speak of the "imaging biobank" and of linking images to genomic and outcome data, turning the archive from a storage cost into a research asset.

AI deployment and algorithm governance

The most recent turning point is the arrival of deep learning in medical imaging in the mid-2010s, which produced a wave of algorithms for detection, segmentation, and triage. The technical history belongs to medical image analysis, but the informatics problem is different and is now central: how does a hospital actually deploy an algorithm safely? Algorithm governance covers the lifecycle—regulatory clearance, integration into the reading workflow, monitoring for performance drift as scanners and populations change, version control, and accountability when an algorithm contributes to a missed diagnosis. Regulators in the United States and Europe created pathways for software as a medical device, and hospitals began building committees and dashboards to track which algorithms run on which patients. The open questions are practical: how to audit a model whose training data is proprietary, how to detect degradation before it harms patients, and who is responsible when a human and an algorithm disagree.

Where the field stands

Imaging informatics today runs on the standards laid down in the 1980s and 1990s—DICOM and HL7 remain the connective tissue—layered with cloud archives, structured reporting templates, and an expanding set of governed algorithms. Its unresolved problems are the ones its history kept producing: interoperability between systems that were never designed to talk, the gap between what a report says and what a computer can use, and the difficulty of proving that a new tool improves outcomes rather than just adding findings. The field's enduring insight is that the image is never just an image; it is a record in a system, and the system is where the clinical value is won or lost.

References

  1. Imaging informatics, Wikipedia
  2. Picture archiving and communication system, Wikipedia
  3. Computer-aided diagnosis, Wikipedia

The frameworks of Imaging Informatics

Each framework is a way of working the field has used: when it held, and the question it answers.

  1. 1970-Present

    Radiological information system

    How does an imaging exam get ordered, tracked and reported?

    Fly to its tower in Nousfarer

  2. 1985-2015

    Computer-aided diagnosis

    Can a computer find the abnormality for the radiologist?

    Fly to its tower in Nousfarer

  3. 1985-Present

    DICOM and HL7 Standards-Based Interoperability

    How do different medical systems exchange images and messages?

    Fly to its tower in Nousfarer

  4. 1985-Present

    PACS and standards-based interoperability

    How can images from many vendors work in one archive?

    Fly to its tower in Nousfarer

  5. 1985-Present

    Picture archiving and communication system

    How are images stored digitally and delivered to the reader?

    Fly to its tower in Nousfarer

  6. 1990-Present

    Workflow and Human-Computer Interaction in Reading

    How should the reading workstation be designed for efficient, safe work?

    Fly to its tower in Nousfarer

  7. 1995-Present

    Structured Reporting and Report-as-Data

    How can a radiology report become searchable data?

    Fly to its tower in Nousfarer

  8. 2005-Present

    Enterprise and cloud image management

    How can a whole health system manage and store its images?

    Fly to its tower in Nousfarer

  9. 2005-Present

    Learning Health System and Imaging Data Science

    How can routine imaging data continuously improve care?

    Fly to its tower in Nousfarer

  10. 2010-Present

    Cloud and Vendor-Neutral Archiving

    How can images be archived independently of any one vendor?

    Fly to its tower in Nousfarer

  11. 2010-Present

    Enterprise Imaging

    How can all of a patient's images live in one governed place?

    Fly to its tower in Nousfarer

  12. 2015-Present

    AI Algorithm Governance

    Who ensures medical AI algorithms are safe and accountable?

    Fly to its tower in Nousfarer

How the ideas connect

Each arrow runs from an earlier idea to the later one it shaped. The list below says what passed between them.

Radiological information systemComputer-aided diagnosisDICOM and HL7 Standards-Based Interop…PACS and standards-based interoperabi…Picture archiving and communication s…Workflow and Human-Computer Interacti…Structured Reporting and Report-as-Da…Enterprise and cloud image managementLearning Health System and Imaging Da…Cloud and Vendor-Neutral ArchivingEnterprise ImagingAI Algorithm Governance

Beyond the field

Where Imaging Informatics's ideas came from in other fields, and what they led to there: the strongest 12 of 49 links.

A city in Nousfarer: a red-carpeted stair climbing terraces to a crowned tower on an acropolis, more towers round it, forest and snowy mountains beyond.
A city in Nousfarer, the game played in the browser.

Imaging Informatics is a planet in Nousfarer

Its city has a tower for each of these 12 frameworks. Land, walk in, and climb one concept at a time.

Land on Imaging Informatics

Related fields

Fields its ideas are linked to

More in Medical Imaging

See where Imaging Informatics sits on the Map of Human Knowledge