Study 1 · Standing dashboard
FDA AI clearances keep rising, and spreading beyond radiology
The FDA has cleared 1,614 AI-enabled medical devices, with a record 335 in 2025. Radiology still leads, but its share of new clearances has fallen from a peak of 87% in 2020 to 75% in 2026.
FDA device list through June 29th 2026
FDA AI clearances hit a record 335 in 2025
FDA AI-enabled devices cleared per year, by review panel; 2026 covers January to June
- Radiology panel
- All other panels
Show the numbers
| Year | Cleared | Radiology panel | Other panels |
|---|---|---|---|
| 2015 | 6 | 1 | 5 |
| 2016 | 18 | 11 | 7 |
| 2017 | 27 | 16 | 11 |
| 2018 | 65 | 40 | 25 |
| 2019 | 80 | 51 | 29 |
| 2020 | 114 | 99 | 15 |
| 2021 | 129 | 110 | 19 |
| 2022 | 162 | 138 | 24 |
| 2023 | 226 | 178 | 48 |
| 2024 | 235 | 183 | 52 |
| 2025 | 335 | 254 | 81 |
| 2026 | 181 | 135 | 46 |
Source: FDA AI-enabled device listHealthcare AI Insights
Radiology's share peaked at 87% in 2020; 2026: 75%
Share of each year's FDA AI clearances reviewed by the radiology panel
Since 2020, clearances from other panels grew 5.4 times to 2025; radiology's grew 2.6 times.
Source: FDA AI-enabled device listHealthcare AI Insights
CT leads, with 446 of 1,614 devices
FDA AI-enabled devices by the kind of image or signal they read
Not shown: 95 devices whose summaries name no image or signal.
Source: FDA AI-enabled devices; Cleared to Paid classifier (modality matches reviewed labels on 83% of sampled devices)Healthcare AI Insights
AI and software companies file 67% of devices
FDA AI-enabled devices by who filed them
Scanner makers file most scanner features; AI companies file most triage and detection software.
Source: FDA AI-enabled devices by parent company; Cleared to Paid classifier (company type matches reviewed labels on 94%)Healthcare AI Insights
The top three's share of detection clearances fell from 57% to 20%
Share of detection and diagnosis (CADe/CADx) clearances held by the three biggest filers, by period
Clearances are not revenue: more filers does not mean more buyers.
Show the numbers
| Period | Devices | Companies | Top-3 share |
|---|---|---|---|
| 2016-2018 | 7 | 6 | 57% |
| 2019-2021 | 21 | 14 | 33% |
| 2022-2024 | 47 | 33 | 23% |
| 2025-2026 | 46 | 35 | 20% |
Source: FDA AI-enabled devices by parent companyHealthcare AI Insights
More summaries name their model; few say where data came from
Share of FDA AI device summaries that disclose each item, 2016–21 vs 2022–26
- 2016–21
- 2022–26
What a summary says, not what the device has: silence is a disclosure gap.
Source: FDA device summaries; Cleared to Paid keyword rules and a sentence-by-sentence reading of training textHealthcare AI Insights
58% of new AI devices come from outside the US
Share of FDA AI clearances filed from outside the US, by period (filer address known)
Median FDA review time: 136 days for a 510(k), 288 for a De Novo.
Source: FDA AI-enabled device list (filer addresses)Healthcare AI Insights
Chinese firms hold 70 clearances; United Imaging alone 45
FDA AI clearances by China-headquartered parent companies (4.3% of all), top five
Mostly scanner features. China's own approvals are counted in the report's China tab.
Source: FDA AI-enabled device list by parent companyHealthcare AI Insights
Looking for one device? Search all 1,614 in the report's device explorer.
What the numbers say (6)
High confidence. FDA AI clearances are rising, not stagnating: 335 in 2025, a record, and 181 in the first half of 2026.
Why this confidence?
Counted from FDA's AI list.
Medium confidence. Two business models: scanner makers file 87% of scanner features; AI companies file 95% of triage and 87% of detection devices.
Why this confidence?
Depends on classifier labels (company type matched the reviewed labels for 94% of sampled devices).
Medium confidence. Detection is fragmenting, not consolidating: the top-3 share of detection clearances fell from 57% (2016-18) to 20% (2025-26).
More, and why this confidence?
Clearances are not revenue.
Classifier labels; clearance counts only.
High confidence. First-mover leads are short: 18 of 40 device types created through De Novo have a second company, after about 2 years (median).
Why this confidence?
Counted from FDA decision dates and parent companies.
High confidence. Big tech is nearly absent from FDA's AI list: no clearances for OpenAI, Anthropic, Amazon, NVIDIA, Epic, Oracle or Meta; Apple 5, Google with Fitbit and Verily 4, Microsoft 1.
More, and why this confidence?
Their clinical AI is sold as non-device software.
Counted from FDA's AI list by parent company.
Medium confidence. FDA summaries disclose more since 2022: naming a model type rose from 35% to 51%…
More, and why this confidence?
…mentioning training data from 17% to 49%, and naming where training data came from (a dataset, institution or country) from 0.2% to 6.9%.
Keyword rules, fixed on 2026-10-13; they match the reviewed labels on 92% of devices for whether training data is mentioned. The source share was checked sentence by sentence; the keyword tags had overstated it (17%).
How the labels were built and how often they are right: the method and its error rates.