By Helathora | Published: October 2026 | Last updated: October 4, 2026
AI is now a practical part of radiology rather than a purely experimental technology. Hospitals use AI to flag potentially urgent findings, prioritize examinations, analyze images and support reporting workflows. But AI assisting a radiologist is not the same as AI replacing a radiologist: most clinical systems are designed for specific tasks inside a workflow where clinicians remain responsible for interpretation and decisions.
Radiology continues to dominate the FDA’s AI/ML device landscape, but that does not mean every radiology AI product has the same evidence, regulatory status or clinical value. The more useful question for a hospital is not simply «which tool is best?» but which tool solves a defined problem in our workflow, with evidence that applies to our setting?
How We Researched This Article
This is a desk-research article. We did not independently test any product. We:
- Reviewed FDA information, official company materials, and peer-reviewed or independent studies.
- Labeled company deployment, reach and performance figures as vendor-reported where they came from manufacturers.
- Checked the original study or company source where possible rather than relying only on secondary summaries.
- Separated regulatory status from clinical evidence: FDA clearance does not by itself establish that a product improves outcomes in every hospital.
- Did not rank the companies. Stroke triage, chest-X-ray analysis, abdominal CT triage and reporting assistance solve different problems.
Leading Radiology AI Tools at a Glance
| Platform | Main focus | How AI is used | Selected vendor-reported figures |
|---|---|---|---|
| Aidoc | Imaging triage | Flags suspected acute findings and prioritizes cases for review, including a comprehensive abdomen CT workflow. | 1,600+ medical centers; 60M+ scans analyzed per year |
| Viz.ai | Stroke and care coordination | Detects potential findings, sends alerts and supports coordination across clinical teams. | 2,000 hospitals; 50+ AI care pathways; 230M patients/lives reported |
| Qure.ai | X-ray and CT | Detects, highlights and categorizes findings, particularly across chest X-ray and CT workflows. | 26 FDA-cleared indications across 9 products |
| Rad AI | Reporting workflow | Assists with report drafting, impressions and follow-up recommendation tracking. | 200+ health organizations |
| Annalise.ai / Harrison.ai | Broad image interpretation | Supports detection and triage across multiple findings on chest X-ray and head CT. | 13 FDA-cleared indications reported by the company |
Important: these figures describe companies or portfolios, not necessarily one individual product. FDA clearance applies to a specific device and intended use, and availability varies by country.
What the Evidence Shows
| Source | Design | Main finding | What it does not prove |
|---|---|---|---|
| JMIR longitudinal analysis (2026) | 1,430 FDA-authorized AI/ML devices through 2025 | Radiology remained dominant; it represented 77.5% of authorizations in the 2023–2025 period. | Authorization counts do not measure patient outcomes or real-world effectiveness. |
| MASAI, Lancet Digital Health (2025) and Lancet (2026) | Large randomized mammography screening trial in Sweden | AI-supported screening detected more cancers and reduced reading workload; interval cancers were 1.55 vs. 1.76 per 1,000 screened women. | This is evidence for a mammography screening workflow, not proof that every radiology AI platform improves care. |
| Yu et al., Nature Medicine (2024) | Reader study involving 140 radiologists and 15 chest X-ray tasks | AI assistance helped some radiologists but harmed others; AI errors could reduce performance. | A reader study does not reproduce every feature of live clinical deployment. |
| Savage et al., AJR (2024) | Prospective evaluation of intracranial-hemorrhage triage | No significant improvement in diagnostic accuracy or report turnaround; positive cases were 147.1 vs. 149.9 minutes. | The result applies to one tool and workflow at one health system and should not be generalized to all AI triage. |
| Batra et al., AJR | Single-center before/after worklist study for PE-positive CTPA | Mean report turnaround fell from 59.9 to 47.6 minutes after AI reprioritization. | It was not randomized, and the result depends on the local workflow and implementation. |
| Aidoc pivotal study (company-reported) | FDA-reviewed study covering 11 new abdomen CT indications | Company reports 97% mean sensitivity and 98% mean specificity across the indications. | These are vendor-reported figures and performance varies by indication. |
Mean report turnaround time (minutes)
Together, the evidence shows why hospital evaluation matters. AI-supported workflows can reduce turnaround time in some settings, while another prospective study found no improvement in a different triage workflow. The strongest randomized evidence in this review comes from mammography screening, a different use case from the platforms discussed above. Vendor performance figures should therefore be treated as a starting point for local validation, not as a guarantee of clinical benefit.
What Can AI Actually Do in Radiology?
- Image analysis. Software can look for specific findings on CT, X-ray or mammography and highlight an area of interest for clinical review.
- Worklist prioritization. Studies can be analyzed in the background and potentially urgent examinations flagged so they can move higher in a reading queue.
- Reporting assistance. Some tools help draft, structure or organize radiology reports and follow-up recommendations. This is a different problem from image detection.
- Clinical workflow and coordination. Some platforms connect detection with alerts, worklists and communication between clinical teams.
Leading Radiology AI Platforms to Know in 2026
Aidoc: broad clinical AI and imaging triage
In January 2026, the FDA cleared Aidoc’s Comprehensive Abdomen CT Triage, powered by its CARE foundation model. The company describes 11 newly cleared indications combined with three previously cleared indications, creating a 14-indication abdomen CT triage workflow. The product is designed to surface potentially acute cases earlier rather than relying on a single-condition algorithm. Aidoc reports more than 1,600 medical centers and more than 60 million scans analyzed per year; these figures are vendor-reported.
Main focus: broad clinical AI, imaging analysis and automated triage.
Viz.ai: from stroke care to care coordination
Viz.ai became known for stroke, where rapid detection and communication can be important. In July 2026, the company reported that its platform was embedded in 2,000 hospitals, with more than 50 AI care pathways and an estimated 230 million patients/lives reached. These are company-reported figures. Its broader proposition is not only finding a potential abnormality but connecting detection with alerts and care coordination.
Main focus: stroke-related AI, clinical alerts and care coordination.
Qure.ai: AI for X-ray and CT
In February 2026, Qure.ai announced FDA 510(k) clearance for qXR-Detect covering six new chest X-ray indications. The company states that its total FDA-cleared indications had reached 26 across nine X-ray and CT products. Its chest X-ray tools are designed to identify, highlight and categorize findings to support clinicians and radiologists.
Main focus: chest X-ray and CT imaging.
Rad AI: the radiologist’s workflow
Rad AI focuses more heavily on the reporting workflow than on image detection. Its products include tools for reporting, impressions and follow-up recommendation tracking. A reporting assistant therefore addresses a different operational problem from a system designed to flag pulmonary embolism or intracranial hemorrhage. Rad AI reports use by more than 200 health organizations. We did not independently verify that figure, and hospitals should verify the regulatory status and intended use of each specific product.
Main focus: reporting and radiology workflow.
Annalise.ai / Harrison.ai: broad image interpretation
Annalise.ai’s radiology products are marketed under Harrison.ai. The company reports 13 FDA-cleared indications, including applications involving chest X-ray and non-contrast head CT, with additional indications added during 2026. Broader coverage does not automatically mean a product is better for every hospital. The relevant questions remain the specific intended use, evidence, local performance and workflow fit.
Main focus: broad medical image analysis.
What Does «FDA Cleared» Actually Mean?
FDA clearance does not mean that an AI system is perfect, autonomous or capable of replacing a physician. Regulatory status applies to a specific device and intended use. The FDA uses different regulatory pathways, including 510(k) clearance, De Novo classification and Premarket Approval. A large proportion of radiology AI devices have entered through the 510(k) pathway, which evaluates substantial equivalence to a legally marketed predicate device.
The FDA’s AI-enabled medical device list is a useful starting point, but the agency notes that the list is not comprehensive. Hospitals should therefore verify the exact product, indication, manufacturer, regulatory pathway and current status before purchase.
Does AI Replace the Radiologist?
For the clinical technologies discussed here, no. A typical assistive workflow is:
An imaging study contains information beyond the specific finding an algorithm was developed to detect. Radiologists also integrate clinical history, previous examinations and other information. Research has shown that AI assistance can help some readers while harming others when the system makes errors, which is one reason human review remains essential.
What Are the Limitations of Radiology AI?
- False positives. Too many alerts can create additional work instead of saving it.
- False negatives. AI can miss findings, so sensitivity, specificity and intended use need to be understood for the actual clinical application.
- Differences between hospitals. Patient populations, scanners, protocols and workflows vary, so performance may differ after deployment.
- Workflow problems. An accurate algorithm may still fail to deliver operational value if alerts are poorly integrated into the worklist.
- Automation bias. Clinicians may be influenced by an incorrect AI output, which makes appropriate oversight important.
- Regulatory and model changes. Product indications, software versions and regulatory status can change, so hospitals need ongoing monitoring rather than a one-time evaluation.
How Should a Hospital Evaluate Radiology AI?
| What to evaluate | Question to ask |
|---|---|
| 🩺 Clinical use | What specific clinical or operational problem are we trying to solve? |
| 🔬 Evidence | Has the product been independently evaluated, ideally prospectively and in a setting similar to ours? |
| 🏥 Workflow | Does it integrate directly with our PACS, RIS, worklist and existing clinical systems? |
| ⚖️ Regulation | What exactly is this specific product authorized or cleared to do? |
| 🔐 Data and security | Where is patient data processed and stored, and what security and privacy controls apply? |
| 👨⚕️ Human oversight | Who reviews the AI output, and what happens when the AI disagrees with the clinician? |
| 📈 Monitoring | How will accuracy, alert volume, turnaround time and clinical impact be monitored after deployment? |
| 💰 Total cost | What are the implementation, integration, licensing, support and ongoing operating costs? |
How Much Does Radiology AI Cost?
Enterprise medical AI is not normally priced like a consumer application, and we found no reliable public price list that can be applied across hospitals. Costs may depend on examination volume, number of sites, clinical modules, integration work, support, contract length and deployment model. A simple figure such as «$20 per scan» can therefore be misleading.
Hospitals should request a quote based on their own volume and ask for the total cost of ownership, including implementation and integration rather than only the software license.
What Could the Future Look Like?
The direction of travel is from isolated algorithms toward broader platforms that can analyze multiple findings and connect detection with workflow. Aidoc’s comprehensive abdomen CT approach is one example of this shift, while platforms such as Viz.ai increasingly combine detection with care coordination.
But broader AI does not automatically mean better AI. Future adoption will depend on clinical evidence, interoperability, cybersecurity, regulatory oversight, clinician trust, cost and measurable patient or operational outcomes.
Limitations of This Review
- Company deployment, reach and performance numbers are vendor-reported and were not independently verified by Helathora.
- Several studies are single-center, use before/after designs or evaluate a different product or use case from the current commercial platforms.
- The strongest randomized evidence discussed here comes from mammography screening and should not be generalized to every radiology AI application.
- We did not independently test any of the products discussed.
- FDA device lists and company portfolios change over time, and regulatory status applies to specific devices and intended uses.
The Bottom Line
AI is already part of modern radiology, but the most useful question for a hospital is not which company has the biggest number of clearances or installations. It is whether a specific tool can solve a defined problem, integrate into the existing workflow, demonstrate credible evidence and deliver measurable value in that hospital.
The safest approach is to define the problem, verify the exact intended use and regulatory status, examine independent evidence, test the workflow locally and monitor performance after deployment. AI should be treated as an assistive clinical technology, not as a substitute for qualified radiologists.
For more background, visit the FDA’s AI-enabled medical device information.
Frequently Asked Questions
Is AI already being used in radiology?
Yes. Radiology remains the dominant clinical specialty in the FDA’s AI/ML device landscape, although adoption and clinical benefit vary substantially by product and setting.
What is AI actually used for in radiology?
Common uses include image analysis, detection of specific findings, worklist prioritization, triage, reporting assistance and clinical care coordination. The exact capability depends on the specific device and intended use.
Is FDA clearance the same as FDA approval?
No. The FDA uses different regulatory pathways, and «cleared,» «approved» and «authorized» have specific meanings. Hospitals should check the regulatory status of the exact device and indication they are considering.
Which radiology AI platform is best?
There is no single best platform for every hospital. The appropriate choice depends on the clinical problem, evidence, regulatory status, integration, infrastructure, cost and local workflow.
Will AI replace radiologists?
Current clinical AI systems are primarily designed to assist with specific tasks rather than replace the radiologist’s complete role. Their future impact will depend on evidence, regulation, workflow integration and clinical practice.
Related Reading
Sources
- U.S. FDA. Artificial Intelligence-Enabled Medical Devices. Accessed October 2026.
- Lee YS, Youn BY. Clinical Specialty Expansion of AI-Enabled and Machine Learning–Enabled Medical Devices Authorized by the US Food and Drug Administration From 1995 to 2025: Longitudinal Content Analysis. J Med Internet Res. 2026;28:e103040.
- Aidoc. Comprehensive Abdomen CT Triage. FDA-cleared product information.
- Viz.ai. 10 years, 2,000 hospitals and 230 million patients. July 28, 2026.
- Qure.ai. Six new FDA clearances for qXR-Detect. February 26, 2026.
- MASAI trial: Lancet Digital Health, 2025; final results, The Lancet, 2026.
- Yu F et al. Heterogeneity and predictors of the effects of AI assistance on radiologists. Nat Med. 2024. doi:10.1038/s41591-024-02850-w.
- Savage CH et al. Prospective Evaluation of Artificial Intelligence Triage of Intracranial Hemorrhage on Noncontrast Head CT Examinations. AJR Am J Roentgenol. 2024;223(5):e2431639. doi:10.2214/AJR.24.31639.
- Batra K et al. Radiologist Worklist Reprioritization Using Artificial Intelligence: Impact on Report Turnaround Times for CTPA Examinations Positive for Acute Pulmonary Embolism. AJR Am J Roentgenol. 2023;221(3):324-333. doi:10.2214/AJR.22.28949.