The FDA’s public list of AI-enabled medical devices has passed 1,200 entries, and roughly three quarters are in radiology. No other specialty comes close. Imaging volumes keep rising, while radiologist supply grows slowly, so health systems look to software that can sort worklists, flag urgent findings and measure structures faster than manual work.
Medical imaging AI companies build that software. Their products read CT, MRI, X-ray, ultrasound and mammography data to detect findings, prioritize cases, quantify disease and reduce reporting time. A few vendors sell single-purpose algorithms, such as stroke detection. Others offer platforms that run dozens of tools inside the radiology workflow.
An important distinction applies to every product. AI assistance is not autonomous diagnosis. Most cleared tools are designed to support a qualified clinician, who remains responsible for interpretation. The 20 companies below were selected for documented products, imaging focus and regulatory history. Authorization, diagnostic accuracy and improved patient outcomes are three different things, and the profiles keep them separate.
How AI Is Transforming Medical Imaging
Most imaging AI relies on deep learning, especially convolutional neural networks and, more recently, transformer-based models. Computer vision tasks fall into several types. Detection finds a possible abnormality. Classification labels it. Segmentation outlines organs or lesions for measurement. Reconstruction models improve image quality, shorten scan time or reduce radiation dose.
Integration determines whether a tool is used. Algorithms connect to picture archiving and communication systems, or PACS, radiology information systems and hospital records through DICOM and HL7 interfaces. A tool that forces radiologists to leave their normal viewer is less likely to be adopted.
Three purposes should be kept apart. Diagnostic support suggests or marks a finding for the reader. Workflow prioritization, also called triage, reorders the worklist so urgent cases are read first. Quantitative analysis measures things like lesion volume, calcium score or lung nodule size. Each has different risk and different evidence needs.
Major Medical Imaging AI Applications
| Specialty | Imaging modality | AI task | Clinical workflow | Relevant outcome |
|---|---|---|---|---|
| Neurology and stroke | CT, CT angiography | Detect hemorrhage and large vessel occlusion | Triage, team alert | Time to treatment |
| Breast imaging | Mammography, tomosynthesis | Lesion detection, risk scoring | Screening read support | Cancer detection, workload |
| Chest and cardiovascular | CT, X-ray, CT angiography | Nodules, embolism, calcium, coronary analysis | Prioritization, quantification | Detection, measurement consistency |
| Oncology | CT, MRI, PET | Tumor segmentation, response measurement | Quantification | Reproducible measurement |
| Ophthalmology | Retinal photos, OCT | Retinopathy detection | Screening | Referral rates |
| Pathology | Whole slide images | Cell and tissue classification | Review support | Diagnostic consistency |
| Musculoskeletal | X-ray | Fracture detection | Emergency read support | Missed fracture rate |
AI Diagnostic Accuracy and Clinical Validation
Performance is described with sensitivity, the share of true cases found, and specificity, the share of non-cases correctly cleared. Positive predictive value and negative predictive value depend on disease prevalence, so a tool that looks excellent in a trial can produce many false alarms in a low-prevalence screening population.
External validation matters. A model trained at one hospital may fail at another because of scanner differences, patient demographics or imaging protocols. Dataset bias can lead to lower performance in groups underrepresented in training data. Real-world monitoring after deployment is necessary because performance can drift.
Human oversight remains essential. In practice, the reader sees the AI output and decides. That interaction can produce automation bias, where users accept incorrect suggestions, and the opposite, where users ignore correct ones.
20 Medical Imaging AI Companies
The list below was chosen for imaging AI products with documented clinical use. Regulatory status can change, and each product’s current clearance and intended use should be checked in the FDA database or the relevant regulator’s register before purchase. Some companies are large imaging manufacturers with AI as one component, while others are specialist software firms.
Regulatory clearance means the FDA or another regulator accepted a product for a defined intended use. It is not proof that the product improves patient outcomes, and clearance covers only the indications on the label. Examples labeled hypothetical illustrate workflows and are not documented deployments. Product lists, ownership and clearances change often, so each should be confirmed in the FDA database or the vendor’s current documentation.
1. Aidoc
Company site: aidoc.com
Aidoc, founded in Israel in 2016, runs an AI operating system, called aiOS, that hosts multiple triage algorithms for CT and X-ray, including intracranial hemorrhage, pulmonary embolism, cervical spine fracture and aortic findings. The company has also announced foundation model based products, which should be checked for specific clearances. Its main users are radiologists, emergency physicians and care coordination teams, and its main selling point is a single platform that integrates with PACS and hospital communication systems.
In a hypothetical workflow, a head CT arrives from the emergency department, the algorithm flags a suspected bleed, the study moves to the top of the radiologist’s worklist and a notification reaches the neurosurgery team. The radiologist still reads the scan and decides what to do. Evidence includes peer reviewed studies of turnaround time improvements at individual hospitals, though results vary and false positives add review burden. A limitation is that triage tools help only when the downstream care pathway can respond faster.
2. Viz.ai
Company site: viz.ai
Viz.ai, founded in 2016, received the first FDA authorization for an AI triage tool in 2018 for detecting suspected large vessel occlusion stroke on CT angiography. It then expanded into cardiology, vascular disease and pulmonary embolism. Its core design connects the algorithm to a mobile app that alerts stroke teams, so specialists can review images on a phone and coordinate transfer. It is aimed at hospital networks with a hub and spoke stroke model.
CMS granted a New Technology Add-on Payment for the stroke product in 2020, an early example of payer support for imaging AI. In a hypothetical case, a rural hospital scans a patient with stroke symptoms, the app alerts a comprehensive stroke center and the teams decide on transfer for thrombectomy. Studies have reported shorter times to treatment in some systems, while effects on long term disability depend on the entire care pathway. Limits include alert fatigue and dependence on high quality CT angiography.
3. RapidAI
Company site: rapidai.com
RapidAI, whose RAPID software was developed at Stanford and commercialized by the company, offers automated analysis of CT perfusion, CT angiography and MRI for stroke, along with tools for aneurysm, pulmonary embolism and vascular conditions. RAPID imaging was used in landmark trials such as DEFUSE 3 and DAWN, which supported treating selected stroke patients later than previously thought possible. That trial history gives it unusual clinical credibility.
Users are stroke neurologists, interventional radiologists and emergency teams. In a hypothetical example, perfusion maps show a salvageable region of brain tissue 14 hours after symptom onset and help the team decide on thrombectomy. The software informs the decision and does not make it. Its limits include the need for specific scan protocols and the fact that different vendors’ perfusion maps may not be interchangeable.
4. Qure.ai
Company site: qure.ai
Qure.ai, based in Mumbai and founded in 2016, develops chest X-ray and head CT algorithms. Its qXR product detects findings such as lung nodules, consolidation and signs of tuberculosis, and qER highlights head CT findings such as hemorrhage and midline shift. The company is strong in lower resource settings and in public health screening, where radiologists are scarce. The World Health Organization recommended computer aided detection software for chest X-ray tuberculosis screening in 2021, which supports this category of use.
In a hypothetical example, a mobile screening van takes chest X-rays in a community, the software flags those suspicious for tuberculosis and sends them for confirmatory testing. The tool prioritizes people for testing. It does not diagnose tuberculosis. Performance varies by population and image quality, so local validation is important. The company has received regulatory clearances in several markets, which should be confirmed for each product.
5. HeartFlow
Company site: heartflow.com
HeartFlow, based in Mountain View, produces software that analyzes coronary CT angiography. Its FFRct analysis, first cleared by the FDA in 2014, estimates blood flow to show whether a narrowing limits supply, and its plaque analysis quantifies atherosclerosis. The company reported about $126 million in revenue for 2024 in its 2025 IPO filing and went public in 2025. The company is a specialist and not a general imaging platform.
Cardiologists use it to decide who needs invasive angiography. In a hypothetical case, a patient with chest pain has a CT angiogram, the analysis suggests the narrowing is not flow limiting and the cardiologist avoids a catheterization. Guidelines now give coronary CT angiography a central role for stable chest pain, which supports the use case. Limits include dependence on good CT quality, cost and the fact that some patients still need invasive testing.
6. Nanox AI
Company site: nanox.vision
Nanox AI, formerly Zebra Medical Vision, came into the Nanox group in 2021. It focuses on opportunistic screening, where algorithms scan routine CT and X-ray studies for conditions unrelated to the exam, such as coronary artery calcium, low bone density, vertebral fractures and fatty liver. The audience includes health systems that want to identify patients for preventive care, and population health teams.
In a hypothetical workflow, a chest CT done for lung screening is also analyzed for coronary calcium, and a flagged patient is referred for cardiovascular risk management. The value is early detection of untreated risk. The concern is overdiagnosis and incidental findings that lead to extra testing. Programs need clear follow up pathways. The parent company is also developing a low cost X-ray source, which is a separate business.
7. Imbio
Company site: imbio.com
Imbio, based in Minneapolis, develops quantitative analysis for chest CT, including lung texture analysis for interstitial lung disease, lung density measurement for emphysema and other tools that turn images into measurements. Its audience is pulmonologists and radiologists who need reproducible quantification. Ownership and product lists have changed over time, so a buyer should confirm current details.
In a hypothetical example, a radiologist reviewing a CT for pulmonary fibrosis uses the software to quantify the extent of fibrotic change and track it against a prior scan. Quantification can reduce variability between readers. Limits include the need for standardized scan protocols and the fact that quantitative scores must be interpreted alongside clinical findings.
8. Annalise.ai
Company site: annalise.ai
Annalise.ai, part of the Sydney based Harrison.ai, develops multi finding algorithms, with a chest X-ray product described by the company as detecting more than 100 findings and a head CT product that flags many acute abnormalities. Unlike single purpose tools, these cover many conditions at once, which suits general radiologists and emergency departments.
In a hypothetical case, an emergency chest X-ray shows a pneumothorax and the software highlights it and prioritizes the study. Multi finding systems can produce many flags, so thresholds and review workflows matter. Clearances differ by product and market, so intended use should be checked. Studies show improved reader performance in some settings, though results come from specific datasets.
9. Gleamer
Company site: gleamer.ai
Gleamer, a Paris company, makes BoneView, an algorithm that detects and localizes fractures on X-rays, plus tools for chest radiographs. It targets emergency departments, orthopedic clinics and radiology groups where missed fractures are a recognized problem. BoneView has received FDA clearance and CE marking for its intended use.
In a hypothetical example, an emergency physician reads an ankle X-ray overnight with no radiologist available, and the software highlights a subtle fracture that the clinician then confirms. Studies have reported improved sensitivity for fracture detection when readers use the tool, though real world impact on missed fracture rates depends on adoption. A limit is that the tool is a reading aid and cannot replace clinical judgment about stability or treatment.
10. DeepTek
Company site: deeptek.ai
DeepTek, based in Pune, India, combines AI tools with a radiology reporting platform and teleradiology services used by hospitals and imaging centers, particularly in India. The platform aims to speed reporting by pre analyzing images and organizing worklists for radiologists. Its relevance is in markets with high imaging volumes and limited radiologist supply.
In a hypothetical workflow, a regional imaging center uploads studies to a cloud platform, algorithms highlight abnormal chest X-rays and a remote radiologist reviews and signs reports. The model depends on remote reading infrastructure as much as on algorithms. Regulatory status varies by market, and a buyer outside India should confirm clearances and evidence for each product.
11. Enlitic
Company site: enlitic.com
Enlitic provides data standardization and workflow tools for imaging departments, including software that normalizes DICOM metadata and helps manage data for AI development and deployment. It is more infrastructure than a diagnostic algorithm, and its users are imaging informatics teams and health systems that run multiple AI tools.
In a hypothetical example, a health system with scanners from several vendors uses the platform to harmonize study descriptions so that AI tools and prior comparisons find the right images. The value comes from reducing integration friction. Limits include the fact that the product does not interpret images itself and that benefits depend on the quality of the existing data environment.
12. Lunit
Company site: lunit.io
Lunit, based in Seoul and listed on the KOSDAQ, develops INSIGHT MMG for mammography, INSIGHT CXR for chest X-ray and pathology tools for cancer research and companion diagnostics. In 2024 it acquired Volpara Health, a New Zealand company known for breast density and risk software, to build a screening portfolio. Its audience includes breast imaging centers and oncology researchers.
In a hypothetical screening workflow, a mammogram is read by a radiologist with the software marking suspicious regions and giving a case score, and breast density measurements guide supplemental imaging discussions. Published reader studies show improved cancer detection in some settings. Limits include variation across populations and the need for prospective evidence on interval cancers.
13. ScreenPoint Medical
Company site: screenpoint-medical.com
ScreenPoint Medical, based in Nijmegen, makes Transpara software for mammography and tomosynthesis. It is notable for its use in the Swedish MASAI randomized trial, which enrolled about 80,000 women. Interim results published in The Lancet Oncology in 2023 found that AI supported screening detected 6.1 cancers per 1,000 women versus 5.1 in standard double reading, and it reduced radiologist screen reading workload by roughly 44%.
In a hypothetical workflow, the software scores each mammogram, low risk exams get a single reading and high risk exams get double reading by radiologists. That is a design that saves time while keeping humans in control. Limits include the need for longer follow up on interval cancers and mortality, and the fact that results in Swedish screening may not transfer to programs with different protocols.
14. iCAD
Company site: icadmed.com
iCAD developed ProFound AI for breast imaging, including detection support for tomosynthesis and a risk assessment tool. RadNet, a large US imaging services operator, acquired iCAD in 2025 in a deal reported at about $103 million, after earlier acquiring other AI assets, so the product now sits inside a provider owned technology group. That ownership may affect availability to outside buyers.
In a hypothetical example, a radiologist reads digital breast tomosynthesis with the software highlighting suspicious areas and providing a case score. The tool is aimed at improving reading efficiency and detection. Limits include false positives, which can lead to extra imaging, and the need to confirm current product support after the acquisition.
15. Subtle Medical
Company site: subtlemedical.com
Subtle Medical, founded by researchers from Stanford, develops deep learning software that enhances images from MRI and PET scanners, allowing faster scans or lower radiotracer doses. Its products, such as SubtleMR and SubtlePET, received FDA clearance for image enhancement. These are reconstruction tools, not diagnostic algorithms.
In a hypothetical case, an MRI department shortens scan time for a patient who cannot lie still for long by using the enhancement software to recover image quality. The benefit is throughput and patient comfort. Limits include the need to confirm that enhanced images do not introduce artifacts that mislead readers, and that image enhancement does not by itself prove improved diagnosis.
16. Rad AI
Company site: radai.com
Rad AI, founded in 2018, applies generative AI to radiology reporting. Its tools draft the impression section of a report from the findings a radiologist dictates and help track follow up recommendations for incidental findings. It has raised significant venture funding, including a round in 2024. Its users are radiologist groups and health systems seeking faster reporting and less repetitive dictation.
In a hypothetical workflow, a radiologist dictates findings for a CT abdomen, the software proposes an impression, and the radiologist edits and signs. The product does not interpret the images. Limits include the risk of subtle errors in generated text, so review before signing is essential, and the need to verify that follow up tracking integrates with referring clinicians.
17. Arterys
Company site: tempus.com
Arterys built a cloud based imaging platform, and in 2017 its cardiac MRI product received what was described as the first FDA clearance for a cloud based deep learning application in medical imaging. It also developed oncology and lung tools. Tempus acquired Arterys in 2022, so current products are part of the Tempus portfolio and the standalone brand may be limited.
In a hypothetical example, a cardiologist uses automated segmentation to quantify ventricular volumes from cardiac MRI, which would otherwise take manual contouring. The advantage is time saved and reproducibility. Buyers should confirm current availability and clearances under Tempus, since the original company no longer operates independently.
18. GE HealthCare
Company site: gehealthcare.com
GE HealthCare, an independent public company since 2023, embeds AI throughout its imaging equipment and software, including deep learning reconstruction such as AIR Recon DL for MRI, Precision DL for PET and Sonic DL for faster MRI, along with workflow and decision support tools on its Edison platform. It has among the largest numbers of AI enabled devices on the FDA’s public list.
In a hypothetical example, a hospital buys a new MRI scanner with deep learning reconstruction that shortens exam times and improves image sharpness without a separate software purchase. The advantage is integration with the scanner and service contract. Limits include vendor lock in and the fact that tools mostly work with the company’s own equipment.
19. Siemens Healthineers
Company site: siemens-healthineers.com
Siemens Healthineers offers AI Rad Companion, a family of tools that automate measurements and highlight findings across chest CT, brain MRI, prostate MRI and other studies, and embeds AI in scanners and the syngo.via viewing platform. It also sells deep learning reconstruction for MRI. The company’s breadth across modalities lets large health systems standardize on one vendor.
In a hypothetical example, a chest CT is automatically processed to measure lung nodules and coronary calcium and the results appear in the radiologist’s viewer. The advantage is workflow integration. Limits include the need to confirm which functions are cleared in each market and the fact that tools work best within the Siemens ecosystem.
20. Philips
Company site: philips.com
Philips integrates AI into imaging systems and informatics, with deep learning reconstruction for MRI, workflow tools and an enterprise imaging platform, and it has promoted cloud based tools for radiology and cardiology. The company’s imaging business is separate from its respiratory device recall problems, which concern sleep apnea devices and not imaging software.
In a hypothetical example, an MRI department uses accelerated imaging software to shorten scans and improve patient throughput, and a radiology group uses an informatics platform to route studies. Strengths include scanner integration and a large installed base. Limits include dependence on the Philips ecosystem and the need to confirm which AI features are cleared for specific uses.
| Company | Platform | Specialty | Modality | Intended use | Regulatory note |
|---|---|---|---|---|---|
| Aidoc | aiOS | Multi-specialty CT | CT | Triage, notification | Multiple FDA clearances |
| Viz.ai | Viz platform | Stroke, cardiology | CT, others | Detection, care coordination | FDA authorized |
| RapidAI | Rapid suite | Stroke, vascular | CT, MRI | Perfusion, triage | FDA cleared |
| Qure.ai | qXR, qER | Chest, head | X-ray, CT | Detection | FDA and CE products |
| HeartFlow | Plaque, FFRct | Cardiology | CT angiography | Analysis | FDA cleared |
| Nanox AI | Population health tools | Opportunistic screening | CT, X-ray | Incidental findings | FDA cleared products |
| Imbio | Lung analysis | Chest | CT | Quantification | FDA cleared |
| Annalise.ai | Annalise suite | Chest, head | X-ray, CT | Multi-finding detection | Regulatory clearances |
| Gleamer | BoneView | Musculoskeletal | X-ray | Fracture detection | FDA and CE |
| DeepTek | DeepTek tools | Radiology reporting | X-ray, CT | Report support | Varies by market |
| Enlitic | Enlitic platform | Data and workflow | Multiple | Standardization | Varies |
| Lunit | INSIGHT MMG, CXR | Breast, chest | Mammography, X-ray | Detection | FDA and CE |
| ScreenPoint | Transpara | Breast | Mammography | Reading support | FDA and CE |
| iCAD | ProFound AI | Breast | Mammography | Detection, risk | FDA cleared |
| Subtle Medical | SubtleMR, SubtlePET | Reconstruction | MRI, PET | Image enhancement | FDA cleared |
| Rad AI | Rad AI Reporting | Radiology reporting | Report text | Impression drafting | Software, not diagnostic |
| Arterys (Tempus) | Cloud tools | Cardiac, oncology | MRI, CT | Quantification | FDA cleared |
| GE HealthCare | Edison and scanners | Multi-specialty | All | Reconstruction, workflow | Many FDA listings |
| Siemens Healthineers | AI-Rad Companion | Multi-specialty | All | Quantification, workflow | Many FDA listings |
| Philips | Imaging AI | Multi-specialty | All | Workflow, reconstruction | Many FDA listings |
AI Imaging Applications Across Clinical Specialties
Brain Imaging and Stroke Detection
Stroke care is time-sensitive, and AI triage targets delay. Algorithms detect hemorrhage on non-contrast CT and large vessel occlusion on CT angiography, then alert specialists on mobile devices. Care coordination features let hospitals share images with stroke centers. Studies report faster notification and shorter door-to-treatment times in some systems. Evidence for better patient disability outcomes is less clear, and benefits depend on the care pathway around the alert.
Breast Imaging and Cancer Detection
Mammography screening reads are high volume, and most cases are normal. AI can flag suspicious regions, assign risk scores and triage normal cases. The MASAI randomized trial in Sweden reported higher cancer detection and a large drop in reading workload with AI-supported screening, and a German study of real-world screening reported higher detection rates as well. Questions remain about interval cancers, false positives and long-term mortality.
Cardiovascular and Chest Imaging
Tools measure coronary calcium, plaque and fractional flow reserve, detect pulmonary embolism and identify lung nodules. Opportunistic screening finds incidental conditions on scans done for other reasons. That creates value but also the risk of overdiagnosis, because not every finding needs treatment.
Oncology and Advanced Image Analysis
AI segments tumors, measures lesion size and tracks response across scans. Quantitative consistency can reduce variability between readers. Radiomics extracts large numbers of imaging features to predict outcomes, though many radiomic models have not been validated in prospective trials.
Comparing Medical Imaging AI Platforms
Buyers compare supported modalities, specialty coverage, deployment options, workflow integration and clinical use. Standalone algorithms are narrow and often accurate for a defined task. Integrated platforms offer many algorithms through one contract and interface, which simplifies governance but may include tools of uneven quality.
Hospital procurement should include technical validation on local data. Local testing checks performance on the hospital’s scanners and patients. Contracts should cover cybersecurity, support, update management and exit options.
| Requirement | What to examine |
|---|---|
| Integration | PACS, RIS and EHR connections, DICOM and HL7 support |
| Cybersecurity | Data handling, access controls, vulnerability management |
| Training | Reader education, workflow design |
| Regulatory documentation | Clearance letters, intended use, version history |
| Monitoring | Performance tracking, drift detection, incident reporting |
| Cost factors | Per-study fees, platform licenses, integration work |
Regulatory Approval, Data Security and Clinical Risks
In the US, most imaging AI reaches the market through FDA 510(k) clearance, with some products authorized through De Novo. Clearance relies on substantial equivalence to a predicate and does not always require prospective clinical outcome trials. The European Union uses CE marking under the Medical Device Regulation, and other regions have their own systems.
Intended use restrictions are important. A tool cleared for triage of a specific finding on a specific scan type may not be used outside that scope. FDA has also developed guidance for predetermined change control plans to let some models update safely.
Risks include algorithm drift, demographic bias, false positives, false negatives, cybersecurity incidents and unclear liability. Clinical governance should include human review, a named owner for each tool and ongoing monitoring after deployment.
The Future of AI Powered Medical Imaging
Foundation models trained on large imaging and text datasets may support multi-task analysis and automated draft reports. Multimodal systems may combine images with laboratory data and records. Longitudinal image analysis could track changes over years.
Obstacles include explainability, prospective validation, reimbursement and clinician adoption. Reimbursement for AI remains limited in many settings, which slows deployment. Research gaps include patient outcomes, effects on health equity and performance across diverse populations. The transition from standalone algorithms toward governed, interoperable imaging intelligence will depend on evidence, and qualified clinicians will remain responsible for diagnosis.
Frequently Asked Questions
Q: What are medical imaging AI companies?
A: They are companies that build software using AI to analyze medical images. Products detect findings, prioritize urgent cases, measure structures and improve image quality. Some are specialists, while others are large imaging manufacturers.
Q: How does AI analyze medical images?
A: Deep learning models trained on large labeled image sets learn patterns linked to findings. The software then flags regions, assigns scores or measures structures on new scans. A clinician reviews the output.
Q: Can AI detect cancer in medical scans?
A: AI can help detect suspicious findings in mammograms, chest images and other scans. Studies show it can improve detection or reduce workload in some settings. It does not diagnose cancer without clinician review and pathology.
Q: Is AI more accurate than radiologists?
A: Results depend on the task and study design. Some tools match or exceed average readers on narrow tasks in studies. Radiologists plus AI often perform best, and real-world results vary.
Q: Which AI medical imaging products have FDA clearance?
A: Many do, including products from Aidoc, Viz.ai, RapidAI, Lunit, iCAD, HeartFlow and the large imaging makers. The FDA publishes a list of AI-enabled devices. Check the intended use for each product.
Q: How does imaging AI integrate with PACS?
A: Integration uses DICOM and HL7 standards. The AI receives studies, analyzes them and returns results to the viewer or worklist. Good integration avoids extra steps for the reader.
Q: What are the risks of AI in radiology?
A: Risks include false positives, missed findings, bias, drift and automation bias. Cybersecurity and liability add concern. Governance and monitoring reduce risk.
Q: What is the future of medical imaging AI?
A: Expect foundation models, multimodal analysis, automated reporting and better longitudinal tools. Adoption depends on prospective evidence, reimbursement and clinician trust.