The phrase "AI companies in healthcare" covers a wide range of organizations — from radiology software vendors with dozens of FDA-cleared devices to early-stage drug discovery platforms that have never submitted to a regulator. Treating them as a single category makes comparison difficult and evaluation nearly impossible.
This reference organizes the landscape by application domain, regulatory status, and deployment context. The goal is to give professionals a structured starting point for verification and comparison — not a ranked list or a buyer's guide.
How to Read This Reference
Before evaluating any AI company in healthcare, it helps to clarify three things: what problem their product addresses, whether the product is classified as a medical device under FDA rules, and what evidence exists beyond the vendor's own materials.
- Application domain: radiology AI, AI scribes, clinical decision support, revenue cycle automation, drug discovery, and pathology AI each have distinct regulatory pathways, evidence standards, and deployment requirements.
- FDA authorization status: Software as a Medical Device (SaMD) requires FDA clearance or approval before clinical deployment in the U.S. Many AI tools marketed to health systems are not SaMD and face no premarket review.
- Evidence quality: FDA clearance confirms a device meets a regulatory standard — it does not confirm clinical utility. Post-market real-world evidence studies are a separate data point and are noted where available.
- Funding stage: Seed and Series A companies may have compelling prototypes but limited deployment history. Public companies have disclosed financials but face commercial pressures that affect product roadmaps.
Application Domains and Representative Company Types
The following table maps the major application domains to their typical regulatory classification, primary deployment setting, and the type of evidence that matters most when evaluating vendors in each category.
| Domain | Typical Regulatory Class | Primary Setting | Key Evidence Type |
|---|---|---|---|
| Radiology AI | SaMD — 510(k) or De Novo | Hospital / outpatient imaging | External validation studies, RWE deployment data |
| AI Scribe / Ambient Documentation | Not typically SaMD | Ambulatory, inpatient | Clinician adoption rates, note accuracy audits |
| Clinical Decision Support (CDS) | SaMD if meets criteria; otherwise non-device | EHR-integrated, ICU, ED | RCT or prospective validation; bias audits |
| Pathology AI | SaMD — 510(k) or De Novo | Surgical pathology labs | Concordance studies, external validation |
| Revenue Cycle / Prior Auth AI | Generally not SaMD | Administrative / billing | Denial rate reduction, audit accuracy |
| Drug Discovery AI | Not SaMD (discovery phase) | Pharma R&D | Pipeline progression, published target validation |
| Cardiology AI | SaMD — varies by use case | Cardiology, primary care | Clinical trial outcomes, sensitivity/specificity |
Radiology AI Companies
Radiology has the highest concentration of FDA-cleared AI devices of any clinical specialty. As of Q2 2026, the FDA's AI/ML-enabled device list includes well over 700 authorized devices, with radiology and cardiology accounting for the majority. This concentration reflects both the structured nature of imaging data and the relatively clear performance metrics available in this domain — sensitivity and specificity for specific findings.
What Radiology AI Companies Actually Do
Most radiology AI products fall into one of three functional categories: triage tools that flag urgent findings (intracranial hemorrhage, pulmonary embolism) for expedited radiologist review; detection aids that identify candidate findings (nodules, fractures, lesions) within a study; and quantification tools that measure anatomical structures or pathological burden.
Companies like Aidoc, Viz.ai, and Avicenna.AI have built multi-condition triage platforms with multiple FDA clearances. Smaller vendors often hold clearance for a single indication. The number of FDA clearances a company holds is a useful proxy for regulatory maturity — but not for clinical utility, which depends on validation study quality and real-world deployment performance.
External Validation: The Persistent Gap
FDA clearance for a radiology AI device typically requires performance data on a test dataset — but that dataset is often drawn from the same distribution as the training data, and it may not reflect the demographics or scanner hardware present in the deploying institution. External validation on independent datasets from different health systems is a stronger signal, and it remains uncommon in the published literature for many cleared devices.
Algorithmic bias concerns are particularly acute in radiology AI. Models trained predominantly on images from large academic medical centers may underperform on scans from community hospitals, rural facilities, or patient populations with different baseline disease prevalence. Several published audits have documented performance gaps by race, sex, and age for FDA-cleared imaging AI tools.
AI Scribe and Ambient Documentation Companies
Ambient AI documentation — software that listens to a clinical encounter and generates a structured note — has seen rapid commercial adoption since 2023. Companies including Nuance (Microsoft), Abridge, Suki, Nabla, and DeepScribe have deployed at scale across major health systems. This is one of the few healthcare AI segments where clinician adoption has been broadly positive, largely because the workflow benefit is immediate and the risk profile is lower than diagnostic AI.
Most ambient documentation tools are not classified as SaMD under current FDA guidance, because they generate documentation for clinician review rather than driving clinical decisions. This means they face no premarket FDA review — which matters when evaluating accuracy claims. Vendors typically report note accuracy metrics from internal testing or small pilot studies, not from independent peer-reviewed validation.
Funding Stage and Deployment Scale
Abridge (Series C, backed by UPMC and others) and Nuance DAX (part of Microsoft, publicly traded parent) represent different ends of the funding spectrum. Abridge has published peer-reviewed deployment data from UPMC; Nuance DAX has broader commercial reach by virtue of Microsoft's existing EHR integrations. Neither funding status nor parent-company size is a reliable proxy for note quality — that requires independent accuracy audits.
Clinical Decision Support Companies
Clinical decision support (CDS) AI sits at the intersection of the highest potential benefit and the highest regulatory complexity. Tools that predict sepsis, flag deteriorating patients, recommend medication adjustments, or stratify surgical risk are the applications where AI could most directly affect outcomes — and where the consequences of false positives and false negatives are most significant.
The Sepsis Prediction Example
Sepsis prediction algorithms have been among the most studied and most criticized CDS AI tools. Epic's Sepsis Prediction Model, deployed across hundreds of health systems, was the subject of a widely cited 2021 JAMA Internal Medicine study that found the model performed substantially worse in external validation than in vendor-reported figures — with a positive predictive value under 10% in some settings. This case became a reference point for why vendor-reported performance figures should not be treated as generalizable.
Companies like Dascena, Philips (with its Early Warning Scoring systems), and newer entrants have continued to develop sepsis and deterioration prediction tools. The evidence base remains mixed. For any CDS AI tool, the relevant questions are: what was the study population, was the model externally validated, and what happened to clinical workflows and alert fatigue when it was deployed?
Regulatory Classification of CDS Tools
FDA's 2023 final guidance on clinical decision support software clarified which CDS tools are subject to device regulation. Tools that are intended to acquire, process, or analyze medical images, signals, or patterns are generally regulated as SaMD. Tools that display or organize patient data for a clinician to independently review are generally not. The boundary is actively contested, and several CDS vendors have received warning letters or been asked to seek clearance for products they had marketed as non-device.
Drug Discovery AI Companies
Drug discovery AI operates almost entirely outside the FDA's SaMD framework — at least in the early pipeline stages. Companies in this space use AI for target identification, molecular generation, binding affinity prediction, and clinical trial design. The relevant regulatory event is the IND application and eventual NDA or BLA filing for the drug itself, not the AI software.
Prominent companies in this space include Recursion Pharmaceuticals (public, NYSE: RXRX), Insilico Medicine, Exscientia (acquired by Recursion in late 2024), Schrödinger, and Absci. Evaluating these companies requires tracking pipeline progression — which compounds have reached Phase I or Phase II trials, and what the attrition rate looks like — rather than FDA device clearance counts.
Revenue Cycle and Administrative AI Companies
Revenue cycle management (RCM) AI and prior authorization automation represent the largest commercial market for healthcare AI by contract volume, even though these tools have no direct clinical function and face no SaMD regulatory pathway. Companies like Waystar, Availity, Olive AI (now largely wound down), and Cohere Health have built products that automate prior authorization decisions, claim scrubbing, denial management, and coding.
The regulatory environment for payer-side AI is shifting. CMS has issued guidance on prior authorization transparency requirements, and there is active Congressional attention to AI-driven claim denials. The relevant compliance framework here is not FDA SaMD guidance but CMS rules, state insurance regulations, and emerging requirements around algorithmic transparency in coverage decisions.
Pathology AI Companies
Digital pathology AI has a smaller but growing base of FDA-cleared devices compared to radiology. The foundational challenge in pathology AI is that whole-slide imaging (WSI) at scale is a relatively recent capability — many pathology labs are still in the process of digitizing workflows, which constrains deployment.
Paige.AI received FDA authorization for its prostate cancer detection algorithm (De Novo, 2021) — one of the first AI-based pathology tools to receive that level of review. PathAI, Proscia, and Hologic's Genius Digital Diagnostics platform (for cervical cytology) are among the other significant players. As with radiology AI, the key evaluation dimensions are external validation status, demographic coverage of training datasets, and real-world concordance with pathologist diagnosis.
Cardiology AI Companies
Cardiology AI spans ECG analysis, echocardiography interpretation, CT angiography assessment, and wearable cardiac monitoring. Companies like iRhythm (FDA-cleared Zio patch), Eko Health (AI-powered stethoscope), Caption Health (acquired by GE HealthCare), and HeartFlow (coronary artery analysis) hold multiple FDA clearances and have published clinical validation data.
HeartFlow's FFRCT analysis — which estimates fractional flow reserve from CT angiography without invasive catheterization — has one of the more robust clinical evidence bases in healthcare AI, including prospective trials. That level of evidence is not typical across the cardiology AI segment.
Evaluating AI Companies in Healthcare: A Practical Framework
When a health system, payer, or investor needs to evaluate an AI company in healthcare, the questions below provide a structured starting point. They are not a checklist for purchase decisions — that requires institutional due diligence beyond what any reference can provide.
- Is the product classified as SaMD? If yes, what is the FDA submission number and clearance pathway? If no, what is the regulatory basis for that classification?
- What is the intended use as stated in the FDA authorization (or, for non-device tools, as stated in the product documentation)? Does the deployed use match the authorized use?
- What peer-reviewed evidence exists? Is it prospective or retrospective? Was the model validated on an external dataset from a different institution?
- What is the demographic composition of the training and validation datasets? Are there published bias audits or subgroup analyses?
- What is the company's funding stage and disclosed funding total? Has the company disclosed any regulatory actions, product withdrawals, or material legal history?
- Are there published real-world deployment reports from independent institutions — not vendor case studies?
Company Profile Records on This Site
Individual company profiles on Healthcare AI Insights record the following structured fields: primary application area, FDA-cleared product count (as of last update), funding stage, founding year, key clinical partnerships, and notable regulatory or legal history. Profiles are updated when material changes occur — new clearances, acquisitions, funding rounds, or product withdrawals.
Profiles do not rank companies against each other. A company with zero FDA clearances may be operating legitimately in a non-SaMD domain; a company with ten clearances may have thin clinical evidence for most of them. The profile records the facts — the evaluation judgment belongs to the reader.