What the research found
A new study from the Harvey L. Neiman Health Policy Institute, published in the American Journal of Neuroradiology on June 24, 2026, looked at how one of radiology's most-watched AI reimbursement experiments actually played out. The team analyzed 2,116 inpatient acute ischemic stroke episodes across 1,076 facilities from October 2020 through December 2023 — the window in which Medicare offered a New Technology Add-on Payment (NTAP) for AI that flags suspected large vessel occlusion.
Adoption grew year over year but stayed low. NTAP-billed AI use peaked at 21% of eligible cases in 2022, then declined in 2023 as the temporary code began to sunset. Where the AI was used followed a clear pattern: it clustered at comprehensive stroke centers and at hospitals with 1,000 or more beds, and in the Stroke Belt. In adjusted models, the odds of NTAP-billed AI use were roughly 6× higher in 2022, about 2× higher for patients in the Stroke Belt, and around 1.5× higher at comprehensive stroke centers.
The most important result is what did not predict access. The researchers found no disparities in NTAP-billed AI use by patient sex, age, or race/ethnicity, and none by stroke severity. Instead, hospitals serving more socioeconomically deprived areas were significantly less likely to use the AI. As lead author Casey Pelzl, MPH, put it, "access to these technologies depends more on where a patient is treated than on their clinical needs." The gap isn't about which patients need AI — it's about which sites can afford to adopt it.
How the payment design gates access
To see why access concentrated, look at how NTAP works. It is a temporary, inpatient-only add-on that pays a hospital above the standard DRG amount — covering up to 65% of the qualifying technology's cost for no more than three years, after which it sunsets. For the stroke-AI tool the study examined, the per-case add-on was capped at roughly $1,040.
That structure quietly selects for a certain kind of provider. Capturing the payment requires inpatient stroke volume, the coding and billing infrastructure to submit NTAP claims correctly, and the capital to buy and integrate the software while the add-on offsets only part of the cost — and only for a few years. Large academic and comprehensive stroke centers clear those bars easily. Community hospitals and outpatient imaging centers often don't: much of their imaging is outpatient (where NTAP doesn't apply at all), their case volumes are thinner, and a temporary add-on is a shaky basis for a capital decision.
The result is a reimbursement mechanism intended to spur adoption that, in practice, concentrates it. When the code disappears, sites that leaned on it are exposed — and sites that never qualified were never in the game. That is how an access gap becomes a disparity: not through any clinical decision, but through the design of a payment.
Capital-gated adoption vs. per-study access
The underlying problem is that AI adoption has been priced like a capital purchase and financed through a narrow, temporary payment. Change the pricing and the access math changes with it. The contrast:
| Capital + NTAP-gated model | Per-study / performance-based model | |
|---|---|---|
| Upfront cost | Hardware + licensing capital outlay | None; cost is per report |
| Who can adopt | Large hospitals with capital & inpatient volume | Any site, including outpatient & community |
| Depends on temporary code | Yes — exposed when NTAP sunsets | No — cost scales with actual use |
| Setting | Inpatient only | Inpatient and outpatient |
| Access tracks | Site resources | Clinical demand |
A per-study or performance-based price turns AI from a line item on a capital budget into a variable cost that rises and falls with volume. That removes the exact barrier the NTAP research exposes: the need for capital and inpatient scale before a site can even start. A smaller provider can adopt AI CT reporting on the same per-report terms as a large one, without betting on a payment code that expires.
Where this fits with how AI CT reporting works
Access is only half the equation; the other half is what the AI actually delivers once a site can use it. AI CT reporting is built so a foundation model produces a structured, comprehensive report draft across findings — not a single narrow flag — and a radiologist reviews and signs every report before it reaches the chart. Priced per study rather than as capital, that model lets a community hospital or outpatient center get the same ready-to-sign, radiologist-reviewed output as a large center. Reimbursement can still evolve; access shouldn't have to wait for it.
Frequently asked questions
What does the Neiman Institute study say about access to radiology AI?
A 2026 study by the Harvey L. Neiman Health Policy Institute, published in the American Journal of Neuroradiology, analyzed 2,116 inpatient acute ischemic stroke episodes across 1,076 facilities from October 2020 through December 2023. It found that use of AI billed through Medicare's New Technology Add-on Payment (NTAP) program peaked at just 21% of eligible cases in 2022 and was concentrated at comprehensive stroke centers and large hospitals, while hospitals serving more socioeconomically deprived areas were significantly less likely to use it. Lead author Casey Pelzl said access depends more on where a patient is treated than on their clinical needs.
What is NTAP and why does it limit who gets radiology AI?
NTAP, the Medicare New Technology Add-on Payment, is a temporary inpatient-only payment that can cover up to 65% of the cost of a qualifying new technology for no more than three years. Because it applies only to inpatient stays and sunsets after a few years, it favors hospitals with inpatient stroke volume, capital, and coding infrastructure — typically large, resource-rich centers. Outpatient imaging centers and community hospitals see little to none of it, so a reimbursement mechanism designed to spur adoption can end up concentrating access.
Did the study find disparities by patient demographics?
No. The researchers found no differences in NTAP-billed AI use across patient sex, age, or race/ethnicity, or across measures of stroke severity. The disparities were driven by facility-level factors — hospital size, comprehensive stroke center designation, region, and area-level socioeconomic deprivation — meaning the gap is about which sites can adopt AI, not which patients clinically need it.
How does per-study or performance-based pricing improve access to AI radiology?
A per-study or performance-based model turns AI from a capital purchase into a variable, per-report cost that scales with volume. It removes the upfront hardware and licensing spend that reimbursement gates like NTAP were meant to offset, so an outpatient imaging center or community hospital can adopt AI CT reporting without depending on a temporary inpatient add-on payment. Access then tracks clinical demand rather than a site’s capital position.
Source: Pelzl C, Sanmartin M, et al., "Patient and Facility Factors Associated with Uptake of NTAP-Billed AI to Identify Suspected LVO in Ischemic Stroke," American Journal of Neuroradiology (2026), doi.org/10.3174/ajnr.A9494, via the Harvey L. Neiman Health Policy Institute, and as reported by Radiology Business and AuntMinnie. Figures are rounded as reported.