Radiologists

Three parts of your work where AI is already doing real lifting, and what stays yours.

• Fluoroscopic screen1896–1930

  Fluoroscopic screen and photographic plate (wet-plate and dry-film X-ray era)

  The first radiologists worked with photographic plates coated in silver bromide or later in barium-lead sulfate, requiring wet darkroom processing. Fluoroscopy -- projecting a live X-ray image onto a fluorescent screen -- allowed real-time imaging but required the radiologist to work in a darkened room with eyes dark-adapted for many minutes before each session. Interpretation was qualitative: a practitioner's skill was primarily the ability to recognize normal from abnormal anatomy in a single-modality shadow image. Radiation doses were massive by modern standards -- protective gloves and aprons were not standardized until the 1920s and lead shielding was only gradually adopted after the deaths of the early martyrs demonstrated cumulative dose effects.

  Effect on the work

  The first generation of radiologists was small -- perhaps more than a thousand practitioners nationally by 1920 -- and self-selected for tolerance of physical risk. Radiation injury substantially shortened the careers and lives of the earliest practitioners, creating persistent demand for trained replacements even as the specialty was professionalizing.

• Film-screen radiography + fluoroscopy1930–1974

  Film-screen radiography + fluoroscopy (film, darkroom, lightbox era)

  The dominant radiology workflow from the 1930s through the 1970s: X-ray film exposed in a cassette, developed in a darkroom, and hung on an illuminated lightbox for interpretation. The radiologist's entire cognitive process was conducted over that lightbox, with physical films stored in manila envelopes in hospital filing systems. Contrast agents (barium sulfate for gastrointestinal studies, iodinated contrast for vascular and urinary imaging) expanded the range of conditions that plain X-ray could visualize. Nuclear medicine as a distinct sub-specialty emerged in the 1950s and 1960s with the introduction of technetium-99m (1964) as a practical radiotracer. By the late 1960s a complete radiology department included fluoroscopy suites, a film processing darkroom, nuclear medicine cameras (gamma cameras), and diagnostic ultrasound units -- all interpreted by radiologists from static film or Polaroid snapshots.

• CT scanner1974–1990

  CT scanner (EMI/Hounsfield 1971 prototype; first US clinical installations 1973-1974)

  Godfrey Hounsfield at EMI Central Research Laboratories performed the first clinical CT scan at Atkinson Morley's Hospital in London on October 1, 1971. At the October 1972 Chicago meeting of the Radiological Society of North America, Jamie Ambrose presented EMI scanner images to 2,000 assembled physicians and received a standing ovation. US hospitals began installing CT scanners in 1973-1974. The impact on radiology was immediate and transformative: CT produced cross-sectional anatomy in three dimensions with tissue-density differentiation that plain X-ray could not approach, replacing invasive pneumoencephalography (air injected into the brain ventricles) and myelography (dye injected into the spinal canal) for many neurological conditions. Hounsfield and Cormack received the 1979 Nobel Prize in Physiology or Medicine. CT created an entirely new category of radiologist expertise and -- more importantly for the specialty's economics -- justified a new, higher-value billing category that drove radiologist income growth throughout the 1980s.

  Effect on the work

  CT expanded the scope and economic value of radiology more than any prior technology. Between 1975 and 1985, the number of CT scanners in the US grew. Radiology department volume and revenues grew proportionally; the specialty that had been modestly compensated relative to procedural specialties became one of the highest-paid in medicine by the late 1980s.

• MRI1982–1995

  MRI (first commercial scanner 1980; clinical deployment widespread by mid-1980s)

  Raymond Damadian's team produced the first human body MRI scan on July 3, 1977 -- a cross-section of an assistant's chest that took five hours to acquire. FONAR Corporation produced the first commercial MRI scanner in 1980; General Electric, Siemens, and Philips entered the market rapidly thereafter. By the mid-1980s, major academic medical centers and large community hospitals were installing MRI units. MRI's impact on radiology was at least as profound as CT: it provided superior soft-tissue contrast (critical for brain, spine, musculoskeletal, and cardiovascular imaging) without ionizing radiation. The specialty now required radiologists with expertise in not one but two fundamentally different physics platforms. MRI created two new economic drivers: a new high-value billing category, and a strong competitive moat around radiologist expertise -- MRI interpretation required deep technical understanding of pulse sequences, artifact recognition, and tissue characterization that general practitioners could not acquire informally.

  Effect on the work

  Between 1995 and 2006, MRI study volumes in the US nearly tripled per year. The same period saw radiologist compensation reach historic highs, driven by both MRI and CT volume growth under a reimbursement system that valued advanced imaging generously.

• PACS1993–2008

  PACS and DICOM (filmless digital radiology; DICOM standard published 1993)

  The concept of a Picture Archiving and Communication System (PACS) was first articulated in 1982; Dr. Andre Duerinckx coined the term "PACS" in 1981. Early PACS deployments appeared at academic centers from the mid-1980s, but the technology that made interoperability possible was the DICOM standard (Digital Imaging and Communications in Medicine), published in 1993 and establishing a universal format for medical image storage and exchange. PACS transformed the radiologist's work environment from a physical space -- a reading room with hanging lightboxes and film jackets -- into a digital workstation. Studies were read from high-resolution monitors, not from film; prior comparisons were retrieved instantly from archive rather than manually tracked from a film library; radiologists could read from any workstation, including remotely. The productivity effect was dramatic: a radiologist who read many studies per day on film could read more than a hundred studies per day on a digital workstation. This throughput increase drove a substantial income boom in the 1990s and early 2000s and simultaneously created the technical infrastructure that teleradiology (and eventually AI-integrated reading) would later use.

  Effect on the work

  PACS deployment contributed directly to the "bubble years" (1993-2008) in radiology compensation. The same technology also created the infrastructure for offshore teleradiology (Nighthawk Radiology, founded 2001) -- US-licensed radiologists in Australia and Switzerland reading after-hours cases for US hospitals -- which was a direct consequence of the radiologist shortage that PACS-enabled productivity gains had deferred but not eliminated.

• Teleradiology2001–2015

  Teleradiology and voice recognition (remote reading; Dragon NaturallySpeaking in radiology)

  Nighthawk Radiology Services, founded by Paul Berger in 2001, stationed US-licensed radiologists in Australia and later Switzerland to provide nighttime and weekend reading for US hospitals at a favorable time-zone offset. The company grew rapidly by addressing a genuine workforce crisis: there were not enough radiologists to staff 24/7 emergency reads at every US hospital, and PACS had made remote reading technically feasible. By 2005, multiple teleradiology companies were operating and the model had spread beyond after-hours coverage to daytime "overflow" reading for high-volume practices. Simultaneously, voice recognition software (Dragon NaturallySpeaking, later Nuance PowerScribe) replaced tape-recorded dictation routed to transcriptionists, reducing the reporting cycle from hours to minutes. These two changes -- remote reading and instant voice-to-text -- rewired the specialty's workflow and economics and created the infrastructure upon which AI-assisted reporting would later be built.

• FDA-cleared AI triage2016–present

  FDA-cleared AI triage and reporting algorithms (Aidoc 2016, Viz.ai 2018, Rad AI 2020+)

  The first wave of FDA-cleared radiology AI tools reached clinical deployment between 2016 and 2021. Aidoc (founded 2016) built a platform of many FDA-cleared algorithms that run continuously on every qualifying CT and MRI in the PACS, flagging critical findings (intracranial hemorrhage, pulmonary embolism, large vessel occlusion, aortic dissection) before the radiologist opens the study. Viz.ai (founded 2018, many FDA-cleared algorithms, 2,500+ hospital deployments) and RapidAI (many FDA-cleared modules, 2,500+ hospitals) addressed stroke and vascular disease triage. Nuance PowerScribe One's AI-powered Smart Reporting and Rad AI Impressions (launched 2020) began automating the routine measurement and impression-writing tasks that had consumed the most time in high-volume reads. By 2025, roughly 40% of US radiology practices had deployed at least one FDA-cleared AI algorithm in routine clinical workflow; radiology accounts for approximately 75% of all FDA AI/ML medical device authorizations. The key structural feature is regulatory: all FDA-cleared radiology AI is classified as decision support -- the radiologist retains diagnostic authority, signs the report, and bears malpractice liability. No FDA-cleared tool operates autonomously without radiologist attestation, and the regulatory pathway for autonomous AI radiologist replacement does not exist.

  Effect on the work

  AI triage reduces time-to-critical-finding notification by 50-80% at deployment sites; AI impression generation recovers more than an hour per radiologist shift; image enhancement AI (Subtle Medical SubtleMR/SubtlePET) enables 50-80% faster scan acquisition. The net effect across deployed sites is radiologist augmentation: more studies interpreted per radiologist-day, not headcount reduction.