FDA-Cleared AI Medical Devices
A browsable list of AI-enabled medical devices authorized for marketing in the United States, organized by medical specialty and clearance route.
The FDA does not "approve" most medical AI. It clears it. That distinction is the single most misread fact in this entire category, and it changes what a clearance actually tells you about a device.
The agency maintains a public list of AI and machine learning enabled medical devices authorized for marketing in the United States. That list passed 950 entries in 2024 and keeps growing, and roughly three quarters of it is radiology. This page tracks a working subset of that landscape: devices that are well documented, in real clinical use, and useful for understanding how the categories differ.
Three routes appear in the table. 510(k) clearance is by far the most common and rests on substantial equivalence to an existing device, not on proof of clinical superiority. De Novo authorization is used for genuinely novel low-to-moderate risk devices with no predicate, and it creates the classification that later 510(k) devices then use as their own predicate. Premarket approval is the highest evidentiary burden and remains rare in this space.
| Device | Manufacturer | Specialty | Route | Year | Function |
|---|---|---|---|---|---|
| AEYE-DS | AEYE Health | Ophthalmology | 510(k) | 2022 | Autonomous diabetic retinopathy screening from a single image |
| Aidoc BriefCase | Aidoc | Radiology | 510(k) | 2018 onward | Flags suspected findings for prioritized review |
| AIR Recon DL | GE HealthCare | Radiology | 510(k) | 2020 | Deep learning MRI reconstruction from raw data |
| Annalise Enterprise CTB | Annalise.ai | Radiology | 510(k) | 2021 onward | Multi-finding triage on non-contrast brain CT |
| Anumana ECG-AI LEF | Anumana | Cardiology | 510(k) | 2022 | Flags likely low ejection fraction from a standard ECG |
| Arterys Cardio AI | Arterys | Radiology | 510(k) | 2017 | Automated ventricular contouring and quantification |
| BrainScope One | BrainScope | Neurology | 510(k) | 2016 onward | Assesses likelihood of structural brain injury after head trauma |
| Canvas Dx | Cognoa | Psychiatry and Behavioral Health | De Novo | 2021 | Aids autism spectrum disorder diagnosis in young children |
| Caption Guidance | Caption Health | Cardiology | De Novo | 2020 | Coaches a clinician to acquire diagnostic-quality echo images |
| Ceribell Clarity | Ceribell | Neurology | 510(k) | 2018 onward | Detects seizure burden on rapid-response EEG |
| Cleerly | Cleerly | Cardiology | 510(k) | 2019 onward | Quantifies coronary plaque volume and composition |
| Critical Care Suite | GE HealthCare | Radiology | 510(k) | 2019 | On-device pneumothorax detection at the point of acquisition |
| DermaSensor | DermaSensor | Dermatology | De Novo | 2024 | Point-of-care assessment of lesions suspicious for skin cancer |
| EarliPoint System | EarliTec Diagnostics | Psychiatry and Behavioral Health | De Novo | 2022 | Eye-tracking based aid for autism diagnosis and assessment |
| EchoGo | Ultromics | Cardiology | 510(k) | 2020 onward | Automated echocardiographic measurement and analysis |
| Eko Low EF | Eko Health | Cardiology | FDA authorized | 2024 | Point-of-care detection of low ejection fraction |
| EyeArt | Eyenuk | Ophthalmology | 510(k) | 2020 | Autonomous diabetic retinopathy screening |
| GI Genius | Medtronic | Gastroenterology | De Novo | 2021 | Marks suspected polyps during live colonoscopy |
| HealthPNX | Nanox AI | Radiology | 510(k) | 2019 | Flags suspected pneumothorax on chest radiographs |
| HeartFlow FFRct Analysis | HeartFlow | Cardiology | FDA authorized | 2014 | Computes coronary fractional flow reserve from CT |
| icobrain | icometrix | Neurology | 510(k) | 2016 onward | Quantifies brain volumes and lesion burden |
| LumineticsCore (IDx-DR) | Digital Diagnostics | Ophthalmology | De Novo | 2018 | Autonomously detects more-than-mild diabetic retinopathy |
| Lunit INSIGHT CXR | Lunit | Radiology | 510(k) | 2021 onward | Multi-finding detection on chest radiographs |
| Lunit INSIGHT MMG | Lunit | Radiology | 510(k) | 2021 onward | Detection and abnormality scoring for mammography |
| Nevisense | SciBase | Dermatology | Premarket approval | 2017 | Assesses melanoma risk in atypical lesions |
| Optellum Virtual Nodule Clinic | Optellum | Radiology | 510(k) | 2021 | Assigns malignancy risk scores to lung nodules |
| OsteoDetect | Imagen Technologies | Radiology | De Novo | 2018 | Detects distal radius fractures on wrist radiographs |
| Paige Prostate | Paige.AI | Pathology | De Novo | 2021 | Flags areas suspicious for prostate cancer on digitized slides |
| Persyst EEG Analysis | Persyst | Neurology | 510(k) | Multiple clearances | Automated seizure and spike detection in EEG |
| ProFound AI | iCAD | Radiology | 510(k) | 2018 | Detects and scores suspicious findings in tomosynthesis |
| Rapid ICH | RapidAI | Radiology | 510(k) | 2019 onward | Flags suspected intracranial hemorrhage on non-contrast head CT |
| Retinal Screening in Primary Care | Multiple manufacturers | Primary Care and Laboratory | Varies | Ongoing | Autonomous retinal screening deployed outside eye clinics |
| Sight OLO | Sight Diagnostics | Primary Care and Laboratory | 510(k) | 2019 | Point-of-care complete blood count from a finger prick |
| SKOUT | Iterative Health | Gastroenterology | 510(k) | 2022 | Real-time computer-aided polyp detection |
| SubtleMR | Subtle Medical | Radiology | 510(k) | 2019 | Denoising and resolution enhancement on acquired MRI |
| SubtlePET | Subtle Medical | Radiology | 510(k) | 2018 | Denoising for reduced-dose or shortened PET acquisitions |
| Tempus ECG-AF | Tempus | Cardiology | FDA authorized | 2024 | Estimates near-term atrial fibrillation risk from ECG |
| Transpara | ScreenPoint Medical | Radiology | 510(k) | 2018 onward | Region-level and exam-level scoring for breast screening |
| TrueFidelity CT | GE HealthCare | Radiology | 510(k) | 2019 | Deep learning CT image reconstruction |
| Us2.v1 | Us2.ai | Cardiology | 510(k) | 2022 | Fully automated echocardiogram reporting |
| Viz LVO | Viz.ai | Radiology | De Novo | 2018 | Detects suspected large vessel occlusion and alerts the stroke team |
What The Distribution Tells You
Radiology dominates for structural reasons, not because imaging is uniquely suited to machine learning. Imaging produces standardized digital output, has a mature predicate ecosystem to claim equivalence against, and has measurable endpoints a submission can be built around. A pathology or psychiatry device has to establish more of that ground itself, which is why authorizations in those specialties are counted in single digits while radiology is counted in hundreds.
What A Clearance Does Not Tell You
It does not tell you the device outperforms a clinician. It does not tell you the device was validated on a population resembling yours. It does not tell you performance holds outside the conditions studied in the submission. Several devices in this table were authorized on retrospective datasets from a small number of sites, and dataset shift is the normal failure mode when that model meets a different scanner, a different protocol, or a different patient mix.
Read the full explainer on the clearance routes for how the submissions actually work, and the evaluation guide for what to ask a vendor once you have established that a device is cleared.