2026-08-07T18:30:00.000Z
AI Diagnostic Market Reimbursement Gap
The U.S. Food and Drug Administration authorized 331 new AI-enabled medical devices in 2025, the most in the agency's history, pushing its cumulative list past 1,400 devices. Fewer than a handful of those devices carry a dedicated Medicare payment code. That gap, not model accuracy or FDA backlog, is the single fact that should reorganize how a hospital CFO or a diagnostics investor reads this market. Everything else in this report, the CAGR, the named vendors, the funding rounds, sits downstream of it.
Radiology built this market's first decade almost entirely on hardware-embedded algorithms sold alongside CT and MRI scanners, a model that answered the reimbursement question by default because the underlying scan was already billable. That default is breaking down as AI diagnostics moves into blood-based sepsis scoring, standalone ECG analysis, and digital pathology, categories where no imaging claim exists to attach a payment to. Kaiso Research's data puts a specific number on how large that transition becomes by 2035. What it does not show on its own, and what this analysis adds, is which side of the coding gap actually captures it.
Radiology Holds 75% of FDA-Cleared AI. Software Captures the Recurring Revenue.
Kaiso Research's primary dataset puts the global AI diagnostic market at USD 1.97 billion in 2025, expanding to USD 14.09 billion by 2035 at a 21.74% compound annual growth rate across the 2026 to 2035 forecast period. Software leads the component mix through algorithm licensing and SaaS deployment, and Services is the fastest-growing component as hospitals absorb the integration work that AI deployment actually requires. Radiology accounts for roughly 75% of FDA-authorized AI diagnostic devices, the deepest concentration anywhere in the market, while oncology is the fastest-growing application segment tracked in the report.
Hospitals and clinics lead end-use adoption, though imaging centers are now the fastest-growing setting as reimbursement pathways widen beyond hospital walls. North America remains the largest region, with Asia-Pacific growing fastest as national digital-health programs build AI diagnostic capacity into systems that historically had none.
Kaiso Research's data isolates the U.S. AI diagnostic segment specifically at USD 424 million in 2024, expanding at a 17.68% CAGR through 2033, a narrower window than the global 2026 to 2035 forecast but the clearest single read on the market's largest national buyer. That is the shape of the opportunity. What determines who actually gets paid for it is a different question entirely, and it is the question the rest of this analysis is built around.
GE, Siemens, and Philips Built AI Into Hardware. Aidoc and Qure.ai Are Unbundling It.
Two competing models for capturing this market have emerged, and Kaiso Research's company coverage shows both scaling at once. GE Healthcare has embedded more than 40 FDA-cleared AI applications directly into its Revolution CT line, with 96 authorized radiology AI systems tied to its hardware install base. Siemens Healthineers holds more than 450 active imaging-AI patents, and its AI-Rad Companion software cuts chest CT report preparation time by up to 74%, a workflow gain a hospital administrator can put directly into a staffing model.
Koninklijke Philips N.V. carries 42 authorized radiology AI systems into the same hardware-embedded category. Each of these three companies is selling AI as a feature of a capital equipment purchase, not as a standalone line item a radiology department has to separately justify. That distinction matters more than any accuracy benchmark.
The second model sells software independent of any scanner. Aidoc operates 13 acute-finding indications spanning neuro and chest radiology and has built aiOS, an enterprise platform running across nearly 2,000 hospitals worldwide, deployed on more than 110 million patient cases. Riverain Technologies and Vuno Inc. compete on FDA-cleared indication count rather than hardware bundling, while NovaSignal Corporation and Zebra Technologies Corp. round out Kaiso's tracked competitive set with more specialized neurovascular and enterprise imaging plays. AliveCor Inc. sits outside both camps entirely, selling a standalone AI-powered ECG device directly into ambulatory and emergency settings without a hospital imaging contract at all.
A third pattern cuts across both models: partnership as market entry. Siemens Healthineers named Qure.ai the winner of its 2024 Startup Award for AI that improves tuberculosis, stroke, and musculoskeletal diagnosis, pairing Siemens' global imaging hardware footprint with an algorithm built for high-tuberculosis-burden markets Siemens alone does not serve.
F. Hoffmann-La Roche Ltd. is pursuing the same logic in digital pathology, pairing whole-slide imaging hardware with AI-aided tumor classification rather than building an oncology diagnostics AI stack from scratch. Neither company needed to out-engineer a startup. Both simply bought or licensed their way past the years it would have taken to catch up.
Radiologist Shortages, Not Rising Adoption, Are Forcing This Growth
Radiologist shortages are not a soft trend. They are the mechanism converting AI diagnostics from a discretionary pilot into a budgeted line item, and the clinical adoption data now backs that claim with numbers rather than sentiment. A 2024 survey of European Society of Radiology members found that 48% of respondents were actively using AI tools in clinical practice, up from roughly 20% when the same survey ran in 2018.
That is not early-adopter behavior. That is a specialty absorbing a tool because the caseload has outgrown the workforce available to read it. Chronic disease burden compounds the same pressure from the demand side, as aging populations across North America, Europe, and Japan drive imaging procedure volumes higher every year, and each additional scan competes for the same finite pool of trained radiologists.
Kaiso Research's data shows oncology as the fastest-growing application segment for exactly this reason. AI-assisted cancer detection in mammography, pulmonary nodule classification, and colorectal polyp detection has reached clinical accuracy comparable to, and in some published studies superior to, unaided radiologist review. Digital pathology is the newest front in the same shortage story, with whole-slide imaging AI replacing manual slide review in laboratory medicine, and Roche's tumor-classification tools signal that pathology, not just radiology, is becoming a structural growth vector rather than a niche application.
Emerging-market deployment adds a third dimension entirely. India's Ayushman Bharat Digital Mission and comparable digital-health infrastructure programs are building AI diagnostic capability into health systems that never had enough specialists to generate a shortage narrative in the first place. In China, Infervision and Deepwise are deploying lung cancer and cardiovascular AI diagnostics at a scale aimed less at replacing existing radiologists and more at reaching regions that never had consistent specialist coverage. They are simply building capacity that did not exist before.
AI Diagnostics Just Left Imaging for Blood Tests and ECGs
The most consequential recent development in this market has nothing to do with radiology. In April 2024, the FDA granted De Novo marketing authorization to Prenosis Inc.'s Sepsis ImmunoScore, the first AI diagnostic device ever authorized for sepsis, a condition the CDC links to hundreds of thousands of U.S. deaths annually. The software scores 22 clinical and biomarker parameters extracted from the electronic health record to flag sepsis risk within 24 hours. Nothing about it touches a scanner.
That expansion beyond imaging is accelerating, not slowing. AliveCor's Kardia 12L, a handheld AI-powered 12-lead ECG system, launched with FDA clearance for 35 cardiac determinations in June 2024. By January 2026, the FDA had cleared an expanded version detecting 39 determinations, including new rhythm and axis-morphology findings, and the device has since identified more than 4,000 cases of myocardial infarction and ischemia across its installed base.
That update landed after Kaiso Research's underlying report data was compiled, and it matters: cardiology AI is now iterating on a roughly annual FDA cycle, the same cadence radiology AI established years earlier. Foundation models are the structural trend underneath both stories. Aidoc's CARE model received what the company describes as the first FDA clearance for a multi-condition, foundation-model-based triage system in clinical imaging, a meaningfully different regulatory category than the single-finding algorithms that built the market's first decade.
Where a single-task tool clears one indication at a time, a foundation model is built to clear many at once. The FDA's Predetermined Change Control Plan pathway, finalized in December 2024, exists precisely to let that kind of continuously updating system stay authorized as it learns, rather than filing a fresh submission for every incremental capability.
Aidoc's Foundation Model Clearance Signals the End of Single-Task AI
Every AI diagnostic device sold today is still classified as Software as a Medical Device, but the engineering underneath that classification has shifted twice in three years. The first shift moved from single-finding detection, a model trained to flag one abnormality, to multi-finding triage across an entire study. The second, now underway, is the move to foundation models trained across imaging modalities and clinical data types at once.
That second shift changes the regulatory math. A narrow algorithm cleared for one finding needs a new submission every time its scope expands, while a foundation model under a Predetermined Change Control Plan can pre-specify how it will change and stay inside a single authorization as it does. Adoption is still early: only 10% of 2025's AI/ML device clearances included a PCCP, evidence that most of the industry has not yet restructured its regulatory strategy around the mechanism, even though it already exists.
Integration complexity remains the unglamorous constraint underneath all of this. Every AI diagnostic tool has to talk to a PACS, an EHR, and a radiology information system that was rarely built with a machine-learning inference layer in mind. It's a less exciting story than foundation models, but it is the one most hospital IT budgets are actually fighting, quarter after quarter, long after the FDA clearance is already in hand.
Hardware Incumbents vs. Platform Challengers: Who Actually Owns the Workflow?
GE Healthcare, Siemens Healthineers, and Philips are not competing with Aidoc and Qure.ai on algorithm accuracy. They are competing on distribution, and for now, distribution is winning. A hospital that buys a Revolution CT scanner gets 40-plus AI applications bundled into the capital purchase, which is a fundamentally easier procurement decision than a separate software contract, a separate security review, and a separate PACS integration project for a single point solution.
Platform challengers are answering with breadth instead of bundling. Aidoc's argument, that a single integration layer managing dozens of AI modules beats a hospital juggling a dozen separate point-solution vendors, is compelling on paper and increasingly funded in practice. Whether that bet pays off depends on something narrower than either side wants to admit: how many of those bundled or platformed algorithms hospitals actually turn on and use, department by department, not how many are technically licensed on a contract.
The clearest analytical read is that consolidation favors whichever side solves distribution fastest, and right now that is the hardware incumbents. Software-only platforms win the argument on paper. Hardware incumbents win the purchase order, and in a hospital capital budget cycle, the purchase order is the only vote that counts.
Aidoc's $150 Million Signals Where Smart Money Is Betting Next
Capital is flowing toward platform breadth over point-solution depth, and the clearest signal is the size of the checks being written into clinical AI infrastructure rather than single-indication algorithms. Aidoc closed a $150 million Series E led by Goldman Sachs Alternatives' Growth Equity division in April 2026, with participation from General Catalyst, SoftBank Vision Fund 2, and NVentures, Nvidia's venture arm, pushing the company's total funding past $500 million and setting up a possible IPO.
The investor list is the real signal here. SoftBank and Nvidia's venture arm are not healthcare specialists; they are infrastructure investors betting that clinical AI platforms will scale the way cloud infrastructure did a decade earlier. That thesis assumes the reimbursement gap closes fast enough to justify hospital-system-wide platform contracts rather than department-by-department pilots, which is precisely the bet this article's central argument says isn't yet proven at scale.
FDA Clears Devices Faster Than CMS Can Write Codes to Pay for Them
On January 7, 2025, the FDA issued draft guidance covering the full lifecycle of AI-enabled device software, from model description and data lineage through post-market monitoring, formalizing expectations that had previously been scattered across narrower documents. It followed a December 2024 final guidance on Predetermined Change Control Plans and a June 2024 document on transparency principles for machine-learning-enabled devices, three separate releases inside seven months.
CMS has not moved at the same pace. As of early 2026, only a small number of AI-specific CPT Category I codes exist, covering narrow use cases such as Digital Diagnostics Inc.'s LumineticsCore, an autonomous retinal disease detection system formerly sold as IDx-DR, and coronary flow-reserve analysis. Neither represents the broad category of standalone diagnostic algorithms the market's growth numbers assume will get paid. More than 1,400 AI-enabled devices have cleared the FDA as of the agency's most recent 2026 database update, and the number of those devices carrying a dedicated Medicare payment pathway is a rounding error by comparison.
Europe is regulating the same technology through a different door, and the timing matters as much as the mechanism. The EU AI Act classifies AI systems that are safety components of MDR-regulated devices, which covers most diagnostic AI, as high-risk under Article 6(1) and Annex I. But as of March 2026, AI-enabled medical devices are still certified exclusively through the existing Medical Device Regulation framework; the AI Act's high-risk obligations for this category are not yet in force, and a proposed Digital Omnibus package would push full compliance out to August 2028. Vendors have more runway than the headline classification suggests, not less.
What Hospital CFOs and AI Vendors Should Actually Do About the Reimbursement Gap
Hospital CFOs and radiology group leaders should stop evaluating AI diagnostic purchases primarily on FDA clearance count. Clearance confirms the software is legal to sell. It says nothing about whether the payer will cover the scan it touches, and that distinction should move to the top of every procurement checklist alongside the clinical validation data itself.
The systems worth prioritizing are the ones already mapped to an existing CPT pathway, the LumineticsCore model of autonomous retinal screening or FFR-derived cardiac analysis, where the reimbursement question is already answered rather than pending. A hospital that buys ahead of the coding cycle is buying a cost center with good intentions attached.
AI diagnostics vendors and their investors face the mirror problem. A foundation model that can clear ten indications is not worth ten times more than a single-indication tool if only one of those ten indications carries a payment code, and valuations built on indication count alone are mispricing that gap. The vendors positioned to win the next five years are not necessarily the ones with the broadest FDA clearance portfolio; they are the ones running structured coding and coverage strategy alongside their regulatory submissions, treating CMS engagement as a parallel workstream rather than a step that starts after clearance lands.
Three Risks the FDA Clearance Count Does Not Capture
Reimbursement lag is the most immediate structural risk, but it is not the only one. A 2024 cross-sectional analysis of that year's FDA-authorized machine-learning devices found that only 45.8% of device summaries reported any demographic breakdown of their validation cohorts, and just 15.5% reported race or ethnicity data at all. A device can clear the FDA with a validation cohort that does not resemble the patient population a hospital actually serves, and current disclosure requirements make that gap hard for a purchasing hospital to catch before deployment.
Integration complexity is the second risk, and it is a cost risk more than a technology risk. PACS, EHR, and radiology information system integration routinely extends AI diagnostic deployment timelines well past the software licensing budget hospitals originally approved, and that gap between the sticker price and the total cost of ownership is where AI diagnostic procurement projects most often stall.
Regulatory fragmentation is the third. A device cleared under the FDA's Predetermined Change Control Plan framework in the United States faces a materially different, and currently less settled, compliance path under the EU AI Act's Annex I high-risk classification once that regime takes full effect. Vendors building for both markets at once are underwriting two different sets of post-market monitoring obligations on two different timelines, and few have priced that dual-compliance cost into current valuations.
By 2035, the $14.09 Billion Question Is Still About Payment, Not Performance
Kaiso Research's forecast carries the AI diagnostic market from USD 1.97 billion in 2025 to USD 14.09 billion by 2035, a 21.74% compound annual growth rate sustained across a full decade. Nothing in the clinical evidence base or the FDA's authorization pace suggests that trajectory is at risk. The open question is distributional, not directional: whether the reimbursement infrastructure catches up fast enough for software vendors to capture that value directly, or whether hardware incumbents keep absorbing it by bundling AI into equipment sales that already have a payment pathway attached. The CPT code count, not the FDA clearance count, is the number worth tracking between now and 2030.
The AI Diagnostic Market Doesn't Have an Innovation Problem
It has a payment problem wearing an innovation story. Every structural driver in this market, radiologist shortage, chronic disease burden, foundation-model capability, is real and independently verifiable. None of it explains why a market authorizing 331 new devices in a single year still runs on a reimbursement framework built for a handful of narrow, single-purpose tools.
The vendors who treat CMS coding as a parallel track to FDA submission, not a sequel to it, are the ones who will convert Kaiso Research's USD 14.09 billion 2035 figure into actual revenue rather than actual inventory. Everyone else is building a more sophisticated version of a product nobody has agreed to pay for yet.
Kaiso Research's full report breaks the 21.74% CAGR down by component, application, and region at a granularity this article does not attempt, and that segmentation is where the reimbursement gap actually opens or closes for a specific product line, a specific hospital network, or a specific national market. For a buyer deciding where to place the next budget cycle or the next funding round, that level of detail is what separates reading a trend from acting on one. The bottleneck is not innovation.
---
About Kaiso Research and Consulting
Kaiso Research and Consulting is a global market intelligence firm publishing 5,000+ research reports across 11+ industry verticals.
[email protected] | +1 872 219 0417
Dhwani Sharma, Lead Industry Analyst, Kaiso Research and Consulting | Covering AI-enabled healthcare and diagnostic imaging markets across North America, Europe, and Asia-Pacific
Published: 2026-07-02 | Report Code: LSHI1151
Market Study: Access the full index or request a complimentary sample directly via the AI Diagnostic Market Size, Trend & Opportunity Analysis Report page
