Converging AI and Human Roles
Healthcare

Most AI-Enabled Medical Devices by FDA Review Panel: Radiology Holds 76% of the List

The FDA's list of AI-enabled medical devices held 1,614 authorizations on September 4, 2026, and 1,230 of them were reviewed by the Radiology panel. Here is every review panel on the list, ranked by device count, with what the count does and doesn't say about the clinicians who work in each field.

AI-enabled devices on the FDA list, by review panel (8 largest of 18 panels)
Radiology1,230
Cardiovascular154
Neurology73
Anesthesiology30
Gastroenterology-Urology27
Hematology22
Pathology10
Ophthalmic10
Source: FDA AI-Enabled Medical Device List, content current as of Sept 4, 2026 (1,614 devices); counts by MetaRoles.
Chart: devices on the FDA's AI-enabled device list for each of the eight review panels with the most entries; bar widths are relative to Radiology (1,230).

As of · Source: U.S. Food and Drug Administration, AI-Enabled Medical Device List (decisions through June 29, 2026); counts are ours1

AI-enabled medical devices on the FDA list by review panel, as of September 4, 2026
Rank Review panel (lead) Devices Relative to #1 Share of list (%) First entry (year) Decided in 2025 Decided in 2026 (to Jun 29) Source
1 Radiology 1,230 76.2 1998 254 135 1
2 Cardiovascular 154 9.5 2008 30 15 1
3 Neurology 73 4.5 2014 19 8 1
4 Anesthesiology 30 1.9 1997 8 4 1
5 Gastroenterology-Urology 27 1.7 2018 8 4 1
6 Hematology 22 1.4 2001 3 2 1
7= Pathology 10 0.6 1995 4 1 1
7= Ophthalmic 10 0.6 2018 1 0 1
9= General and Plastic Surgery 9 0.6 2011 2 2 1
9= Clinical Chemistry 9 0.6 2015 0 0 1
11 Orthopedic 8 0.5 2023 3 3 1
12= Dental 7 0.4 2018 0 2 1
12= Microbiology 7 0.4 2013 1 1 1
14 General Hospital 6 0.4 2016 0 2 1
15= Obstetrics and Gynecology 5 0.3 2006 2 1 1
15= Clinical Toxicology 5 0.3 2005 0 0 1
17= Ear Nose & Throat 1 0.1 2026 0 1 1
17= Immunology 1 0.1 2017 0 0 1

Metric: the number of devices on the FDA's AI-Enabled Medical Device List whose "Panel (Lead)" column names that FDA review panel. All 1,614 rows of the list were counted, by us, and all 18 panels that appear on it are shown, so the table is exhaustive for the list and not trimmed. A tied rank is marked "=". "Decided" counts use the list's Date of Final Decision. FDA spells one panel "Clinical Toxcicology" on the list; we correct the spelling. The review panel is the FDA group that led the premarket review, not the medical specialty of the clinicians who use the device (see "What a panel is, and isn't"). The list is not comprehensive: the FDA says it identified devices mainly by AI-related terms in their marketing authorization summaries or classification, and that devices whose decision summaries weren't yet published will appear in later updates, so recent months are undercounted.1 The list holds authorizations, not deployments, so these rows carry no evidence grade.

The top 10

#1

Radiology

1,230 devices

The Radiology panel leads the list with 1,230 devices, 76.2% of all 1,614, and decided 254 of the 335 devices on the list with a 2025 decision (75.8%) and 135 of 181 so far in 2026.1 Its 1,230 devices come from 551 distinct company names as written on the list; the most frequent are Siemens Medical Solutions USA (56), Canon Medical Systems (42) and Aidoc Medical (30), counting spelling variants separately.1 The panel's earliest entries are two 1998 premarket approvals, Hologic's M1000 ImageChecker (June 26, 1998) and Carestream Dental's Logicon Caries Detector (September 2, 1998), both premarket approvals.14 The panel also reviews devices that are not image-reading software: the list's most recent Radiology entries on June 29, 2026 include a Philips diagnostic ultrasound system, so a Radiology-panel count is not a count of tools that read scans for radiologists.1 In the O*NET occupation database, radiologists are SOC 29-1224.2

#2

Cardiovascular

154 devices

Cardiovascular is second with 154 devices (9.5%), the first listed in 2008: OBS Medical's Visensia, decided July 17, 2008.1 Of the 154, 30 were decided in 2025 and 15 in 2026 so far; the most frequent company name is Medtronic (7), followed by iRhythm Technologies and AliveCor (6 each).1 The list doesn't say how many of these are used by cardiologists (SOC 29-1212) rather than by technologists or patients.12

#3

Neurology

73 devices

Neurology has 73 devices (4.5%). Its first entry is BrainScope's Ahead 100, authorized November 17, 2014 through the De Novo pathway; 19 were decided in 2025, up from 1 in 2020.13 The most frequent company name is Brainlab (6).1

#4

Anesthesiology

30 devices

Anesthesiology has 30 devices (1.9%). The panel's first entry, Compumedics' sleep system, was decided March 11, 1997; the most frequent company name is Huxley Medical (4).1

#5

Gastroenterology-Urology

27 devices

Gastroenterology-Urology has 27 devices (1.7%). Its first entry is a surgical system, TransEnterix's Senhance, decided May 25, 2018; the polyp-detection software GI Genius from Cosmo Artificial Intelligence followed on April 9, 2021 through De Novo.13 Cosmo's four entries make it the most frequent company name in the panel.1

#6

Hematology

22 devices

Hematology has 22 devices (1.4%), the earliest from 2001, a CellaVision automatic hematology analyzer (March 15, 2001). CellaVision appears four times, the most of any name in the panel.1 Only 5 of its devices were decided in 2025 or 2026, a slower pace than the panels above it.1

#7=

Pathology

10 devices

Pathology and Ophthalmic tie with 10 devices each (0.6%). Pathology's first entry is the oldest on the whole list: BD's AutoPap 300 QC automatic Pap screener, a premarket approval decided September 29, 1995; the panel's other early entry, Neuromedical Systems' PAPNET, followed on November 8, 1995.14 Ophthalmic's first entry is IDx-DR, authorized through De Novo on April 11, 2018, and the panel's other entries include EyeArt (Eyenuk) and AEYE-DS (AEYE Health).13

#9=

General and Plastic Surgery

9 devices

General and Plastic Surgery and Clinical Chemistry tie with 9 devices each (0.6%). General and Plastic Surgery's first entry is Strata Skin Sciences' MelaFind, a premarket approval decided November 1, 2011; Clinical Chemistry's first is the Dario Blood Glucose Monitoring System, decided December 21, 2015, and its latest is from September 8, 2023, so the panel has had no entry since.14 Counts this small are easily moved by a single update to the list.1

#11

Orthopedic

8 devices

Orthopedic has 8 devices (0.5%), every one decided since December 20, 2023, when Enhatch's patient-specific knee instrumentation was authorized; three were decided in 2025 and three in 2026 so far.1 The panel's entries include Precision AI's Surgical Planning System, decided August 5, 2024.1

#12=

Dental

7 devices

Dental and Microbiology tie with 7 devices each (0.4%). Dental's first entry is Dentsply Sirona's CEREC Ortho Software (January 11, 2018), and it has had two decisions in 2026; Microbiology's first is from August 21, 2013, with its latest decided January 22, 2026.1

What a panel is, and isn't

The FDA's list gives each device a "Panel (Lead)": the FDA review panel that led its premarket review. The agency describes the list as identifying AI-enabled devices authorized for marketing in the United States, and each row carries a submission number whose prefix shows the route: K for a 510(k) clearance, DEN for a De Novo classification and P for a premarket approval.1345 By that prefix, 1,553 of the 1,614 devices (96.2%) are 510(k) clearances, 40 are De Novo and 21 are premarket approvals. A 510(k) shows a device is substantially equivalent to one already legally sold; De Novo is a pathway for novel devices that have no legally marketed predicate; premarket approval is the review for Class III devices, the category that includes devices that support or sustain human life.345

Because the panel follows the product's regulatory classification, not the profession that uses it, the counts are a measure of what the FDA has authorized, not of how many clinicians in a field work alongside AI. Gastroenterology-Urology's first entry is a surgical system, and Radiology's latest includes an ultrasound system, as above. A device may also be sold as software as a medical device or be built into hardware, and the list has no column that separates the two.16

How fast the list has grown

The list's earliest decisions are from 1995, but 1,572 of its 1,614 devices (97.4%) have a final decision date of 2016 or later, and the yearly count has risen from 18 in 2016 to 335 in 2025.1 Radiology's share of each year's decisions has stayed between 59.3% (2017) and 86.8% (2020) since 2016, and has eased from that 2020 peak to 75.8% in 2025, so the panel's lead isn't an artifact of the early years.1

As of · Source: FDA AI-Enabled Medical Device List; counts are ours1

Devices on the FDA AI-enabled list by year of final decision, all panels and Radiology, as of September 4, 2026
Year of final decision All panels Radiology panel Radiology share (%)
Before 2016421535.7
2016181161.1
2017271659.3
2018654061.5
2019805163.8
20201149986.8
202112911085.3
202216213885.2
202322617878.8
202423518377.9
202533525475.8
2026 (to Jun 29)18113574.6

Counted by us from the list's Date of Final Decision column. Each row is a count of rows on the list, so the "Before 2016" row covers every decision from 1995 to 2015. 2026 runs only to the list's latest decision, June 29, 2026, and recent years are likely to rise as the FDA adds devices whose decision summaries weren't yet published.1

What the list does not show about jobs

An authorization is a regulator's decision about a device. It doesn't say that any hospital has bought the device, how often clinicians use it, whether it performs a task on its own or helps a person do it, or what happened to staffing, so we do not grade these rows on our evidence ladder: that takes an operator or independent source describing live use. The counts are also not exposure or displacement measures; a field with many authorized devices is a field with many regulated products. For a role-by-role reading of what AI does and with what evidence, see the posts that grade individual cases, such as how an ambient AI scribe works, and how we measure how much of a role AI does.1

Sources

  1. U.S. Food and Drug Administration, Artificial Intelligence-Enabled Medical Devices (the AI-Enabled Medical Device List; all 1,614 rows read from the page's table; content current as of 09/04/2026), retrieved 2026-10-14.
  2. O*NET OnLine (U.S. Department of Labor), Summary Report for 29-1224.00, Radiologists, and 29-1212.00, Cardiologists, retrieved 2026-10-14.
  3. U.S. Food and Drug Administration, De Novo Classification Request, content current as of 09/30/2025, retrieved 2026-10-14.
  4. U.S. Food and Drug Administration, Premarket Approval (PMA), content current as of 05/16/2019, retrieved 2026-10-14.
  5. U.S. Food and Drug Administration, Premarket Notification 510(k), content current as of 08/22/2024, retrieved 2026-10-14.
  6. U.S. Food and Drug Administration, Software as a Medical Device (SaMD), retrieved 2026-10-14.