Seven findings

Each finding carries the confidence rating assigned in the paper’s integration rules. High confidence required official national data plus at least one independent peer-reviewed source. Moderate confidence required a national model with assumption sensitivity or triangulation across non-equivalent sources. Exploratory findings rest on a small survey, modeled scenarios, or indirect capacity proxies.

Robust and nonrobust conclusions

Select a conclusion to read why the evidence does or does not support it.

Scope, research questions, and definitions

The unit of concern is effective diagnostic capacity, not a single national headcount.

Demand changes with population size, age structure, clinical pathways, technology, and physician ordering behavior. Capacity changes with staffing, work patterns, case complexity, infrastructure, geography, subspecialty fit, and retention. The European evidence frames a valid systems question, but the U.S. answer requires different measures because radiology is delivered through a fragmented mix of hospitals, academic medical centers, private groups, national teleradiology organizations, ambulatory centers, and federal systems. Medicare payment, residency caps, state labor markets, and local credentialing create constraints that do not map cleanly onto the National Health Service.

Six research questions

Operational definitions and the non-equivalence to avoid

Every construct in this research carries a companion warning. The warnings matter more than the definitions, because most workforce debate collapses when two parties use the same word for two different measures.

Methods

A convergent mixed-methods secondary evidence synthesis with reproducible descriptive calculations.

Inclusion and exclusion criteria by domain

Population aging and structural demand

The demographic base exposed to age-associated imaging grew much faster than the strict training stock.

Growth of the U.S. population age 65 and older

Interactive rebuild of Figure 2. Source: U.S. Census Bureau annual estimates for 2014, 2015, and 2024, plus the 2020 decennial count. The 2020 point is a census count, not an annual estimate.

View the published figure

Why aging matters beyond the count

The prevalence of cancer, stroke, cardiovascular disease, degenerative musculoskeletal disease, neurocognitive disorders, and acute multimorbidity increases with age. Older patients also generate more longitudinal comparison, surveillance, incidental findings, contrast-risk assessment, and coordination across care settings. A stable examination count can therefore require more interpretive and communication effort.

Population growth alone does not specify the number of appropriate examinations. It establishes a structural exposure base. The demand realized by the system depends on coverage, clinical guidelines, screening policy, technology, care-seeking, substitution across modalities, and ordering behavior. That is why the analysis separates population expansion from per-person and per-encounter imaging intensity.

Imaging intensity among older adults

CT growth is the clearest repeated signal across non-equivalent cohorts and denominators.

Imaging examinations among older adults, 2000 versus 2016

Interactive rebuild of Figure 3. Source: Smith-Bindman et al. (2019), JAMA. Rates are per 1,000 person-years in seven U.S. health systems covering 135.8 million examinations. Endpoint changes: CT +109.8%, MRI +124.2%, ultrasound +52.8%, nuclear medicine -31.9%.

View the published figure

Growth did not plateau uniformly. In later study segments, CT among older adults increased by 5.2% annually from 2014 to 2016, MRI increased by 2.2% annually from 2005 to 2016, and all imaging increased by 3.0% annually from 2014 to 2016. These are within-cohort rates, not current national estimates. Their value is mechanistic. The system experienced rising imaging per person as the population aged.

Medicare fee-for-service emergency imaging intensity, 2013 versus 2023

Per 100 beneficiariesPer 100 ED encounters

Interactive rebuild of Figure 4. Source: Rosenkrantz and Cummings (2025), Radiology. The two panels use different denominators and must not be read as a single series. Emergency encounters per 100 beneficiaries fell from 65.0 to 54.5 while CT per 100 beneficiaries rose from 18.7 to 36.5, a 95.2% increase.

View the published figure

Why intensity can increase

Projection to 2055 and its sensitivity

Holding 2022 per-person utilization constant, a national projection estimated that population growth and aging would increase 2055 imaging demand relative to 2023 by 25.1% for CT, 16.9% for MRI, 26.9% for nuclear medicine, 17.3% for ultrasound, 17.8% for radiography, and 22.5% for interventional radiology. Population growth accounted for 73% to 88% of the modeled increase, and aging accounted for 12% to 27%, depending on modality.

The constant-rate scenario is not a prediction that intensity will remain constant. When recent 2018 to 2022 utilization trends were extended through 2030, the modality range was -5.6% to +45.2%, with CT at +45.2%. Extending trends through 2035 produced CT growth of +59.3% and nuclear medicine decline of -21%. This sensitivity demonstrates why national planning needs modality-specific scenarios rather than a single examination-growth rate.

Radiologist workforce: stock, distribution, and churn

National headcount growth has not eliminated modeled shortfall or operational fragility.

Change in annual radiologist turnover and attrition

Interactive rebuild of Figure 5. Sources: Parikh et al. (2026), JACR for turnover among 39,439 radiologists, and Christensen et al. (2026), AJR for attrition among 41,432 radiologists. Only the endpoint values are published, so each series is drawn as a two-point slope. Definitions and cohort periods differ and the two series are not additive.

View the published figure

Who churns, and where

Turnover showed a U-shaped relationship with workload, increasing above an estimated inflection point. Female radiologists and metropolitan practice were associated with modestly higher odds, while academic practice was associated with lower odds than nonacademic practice. Adjusted attrition odds were higher for subspecialists versus generalists, female versus male radiologists, Midwest versus Northeast practice, nonacademic versus academic practice, and rural-site practice. These are associations from administrative cohorts and should not be read as individual causal explanations or used to characterize any group.

Consolidation and organizational capacity

Average radiologists per practice rose from 9.7 to 17.9 between 2014 and 2023 while the number of radiologists increased. Consolidation may permit subspecialty coverage, common worklists, overnight distribution, and capital investment. It may also distance staffing decisions from local institutions, and it does not, in itself, demonstrate improved access, lower backlog, or better retention.

Subspecialty and distribution mismatch

An ACR Health Policy Institute update described the national baseline imbalance as likely to remain fairly static rather than automatically widen. That view is compatible with, not contradictory to, the HRSA projected shortfall. A deficit can persist while national growth rates broadly track. The operational concern shifts to where supply is located, which services it covers, and whether it is available when needed.

Pediatric radiology illustrates the mismatch precisely. A national headcount can rise while a high-need subspecialty contracts.

Short-term capacity strategies and their guardrails

The residency pipeline, 2014 to 2025

Program count expanded modestly. Strict diagnostic radiology resident stock was nearly flat.

ACGME diagnostic radiology program and resident counts

Interactive rebuild of Figure 6 using the full Appendix B series. Source: ACGME Data Resource Books, 2018-19, 2021-22, and 2024-25 editions. Programs are plotted on the left axis and residents on the right. The resident series declined to 4,551 in 2019-20 before recovering to 4,770.

View the published figure

Why program and resident counts diverge

Program count can rise without a proportionate increase in resident stock when new programs begin, when programs mature below complement, when existing programs change approved complement, when positions go unfilled, or when specialty pathways are reclassified. A snapshot of programs alone would miss the 2019-20 trough and the recovery that followed.

Contextual growth index across demographic, workforce, and training measures

Interactive rebuild of Figure 7, start year equal to 100. Source: Author calculations from Census, ACGME, and CMS-derived national radiologist counts. End years differ across the four measures, which is why the index is contextual rather than causal.

View the published figure

The index shows that the 65-and-older population grew faster than Medicare-enrolled radiologist headcount, accredited program count, and strict DR resident stock over nearby decade-scale windows. It does not estimate how much radiologist capacity is required per older adult, nor does it adjust for productivity, hours, integrated IR, technology, or care setting.

Current program infrastructure

The 2025-26 ACGME specialty update listed 201 diagnostic radiology programs with 5,689 approved positions and 4,868 filled positions, and 106 integrated IR programs with 1,264 approved and 858 filled. The apparent difference between approved and filled counts must not be labeled a vacancy total without program-level validation, because complement, training-year structure, timing, and reporting conventions can differ.

Training lag and net capacity

A residency expansion is a delayed intervention. A new position requires recruitment, clinical volume, supervision, evaluation, and didactic infrastructure, and several years pass before independent practice. Fellowship training can extend the lag. Net workforce gain is further reduced by retirement, attrition, nonclinical roles, part-time practice, and migration across regions. Training policy should therefore be evaluated as a cohort-flow model rather than a count of approved positions.

Teaching-site perspectives on expansion and contraction

Willingness appears greater than realized expansion. Institutional production constraints explain the gap.

Published program-leader perspectives on expansion

Interactive rebuild of Figure 8. Source: Begum et al. (2026), JACR. Survey n = 47 diagnostic radiology program directors or associate program directors reporting a mean of 8.5 positions per entering year. Different denominators apply to each bar: 85% and 27.7% are of 47 respondents, while 84.6% is of the 13 who requested an increase.

View the published figure

Among respondents who had not requested expansion, published barriers included funding, lack of institutional or graduate medical education support, residency caps, hospital denials, resource constraints, faculty-to-resident ratios, time constraints, and accreditation concerns. The survey does not establish that 85% of all U.S. programs can safely expand. The response rate was low, and leaders interested in expansion may have been more likely to participate.

Accreditation triangulation

The 2025-26 ACGME specialty update identified recurring diagnostic radiology citation themes in 2024-25: learning environment; inadequate numbers, time, or development of faculty or core faculty; facilities and equipment; supervision; and evaluation. These categories independently support the survey’s account of binding production constraints. They also warn that increasing complement without parallel investment can threaten educational quality.

Funding and the Medicare GME cap

These positions are not radiology-specific. Federal slots therefore ease but do not resolve the institutional business case for a new diagnostic radiology position.

Framework synthesis: why expansion does or does not occur

Integrated scenarios through 2055

The national balance is assumption-sensitive. Local capacity risk is more robust.

National imaging-demand and radiologist-supply growth scenarios to 2055

Demand by modalitySupply scenarios

Interactive rebuild of Figure 9. Source: Christensen et al. (2025), JACR demand and supply projection studies. Demand assumes constant 2022 per-person utilization. The two panels share a percentage axis but not a common service-standard denominator, so similar percentage growth cannot be read as balance.

View the published figure

Five system scenarios

Select a scenario to read the demand and capacity conditions that define it, the signal the system would show, and the priority response.

What a national gap means to the people inside it

None of these outcomes is measured by headcount. The next generation of U.S. workforce research should anchor demand to explicit service standards: clinically appropriate examinations completed, interpreted, communicated, and closed within defined time windows, stratified by geography, site type, modality, acuity, and patient group.

How the bands are defined

Policy, operating, and research agenda

Build a portfolio that reduces avoidable demand, protects current capacity, and expands training where readiness is demonstrable.

Figure 10 · Author-developed roadmap

Five horizons, sequenced by what each one can actually change

The roadmap is ordered by lag, not by importance. Protection and measurement come first because they are the only actions that change capacity inside two years. Financing and data reform come last not because they matter least, but because they are the slowest to convert into interpreted output.

Source: Author synthesis.

Decision principles

• Optimize for safe, timely, equitable interpreted output rather than examinations or reports per radiologist alone.
• Fund retention and faculty time as capacity interventions, not as secondary wellness benefits.
• Add residency complement only with documented educational readiness and a pathway to geographic or subspecialty need.
• Measure demand management by patient outcomes and missed-diagnosis safeguards, not volume reduction alone.
• Treat AI as a sociotechnical intervention that must generate measurable net time or quality benefit after oversight cost.
• Use national models for planning ranges and local operational data for staffing decisions.

Prioritized intervention portfolio

Filter by horizon to see what is actionable inside a given planning window.

Minimum viable national dashboard

The paper’s research answer is a national diagnostic-capacity observatory. These are the seven domains it would have to carry, with the stratifiers that make each domain interpretable.

Governance for AI and automation

AI may support demand management, worklist orchestration, protocoling, segmentation and measurement, comparison with priors, report drafting, communication, and follow-up tracking. The evidence base supports cautious optimism rather than an assumed productivity dividend. Every deployment should specify the task, baseline workflow, intended benefit, human review, failure modes, subgroup performance, drift monitoring, downtime plan, and whether time saved is actually converted into clinical or educational capacity.

Conclusions

The evidence supports four central conclusions. First, the population most at risk of age-associated imaging is expanding rapidly. Second, imaging intensity, particularly CT, has risen in older-adult and Medicare acute-care cohorts, so demographic growth understates potential workload. Third, accredited training has expanded slowly under the strict diagnostic radiology definition, while the newer integrated IR pathway complicates comparisons across decades. Fourth, national headcount projections do not settle the service question, because churn, work patterns, geography, subspecialty, infrastructure, and service standards determine effective capacity.

The bluntest defensible U.S. statistic is the HRSA modeled 90% radiology adequacy in 2038. It indicates a persistent shortfall under baseline assumptions. It should not be converted into a claim that every market is 10% short or that the national gap must widen through 2055. Long-range studies show that supply and demand growth can overlap, while ACR workforce research emphasizes persistent national imbalance and sharper distributional problems. These findings are compatible when definitions are respected.

Teaching sites appear to have meaningful expansion interest, but interest is not deployable capacity. Training expansion should be selective, funded, phased, and linked to measurable geographic or subspecialty need. Counting new slots without factoring in faculty, case mix, and graduate retention yields an attractive input metric without a service outcome.

The practical answer to patients waiting inside the gap is a portfolio: protect the existing workforce, reduce avoidable demand, redesign noninterpretive work, establish regional coverage and teaching networks, expand training where the education system can support it, and validate automation against patient-level outcomes. Without a linked denominator, the United States will continue to debate shortage percentages while patients experience the time gap.

Further questions

• How many case-mix-adjusted radiologist hours are required per 1,000 older adults by modality and care setting?
• Which components of CT growth reflect high-value substitution, and which reflect duplicative or low-yield pathways?
• How much clinical capacity is lost to turnover, onboarding, credentialing, and unfilled schedules rather than permanent exit?
• Which residency expansion models produce durable rural, pediatric, breast, neuro, emergency, or interventional coverage?
• What faculty-to-resident and protected-time thresholds preserve education while programs expand?
• Which automation tasks produce verified net capacity after review and exception-management cost?
• What service standard should define adequate imaging access nationally, and how should it vary by acuity?

Appendices

Appendix A. Data dictionary and calculation register

Calculation cautions

• CAGR is an endpoint summary and does not imply smooth annual change.
• Percent increases from small bases can appear large. Absolute counts are shown wherever possible.
• Different endpoint years in the contextual index prevent causal or balance interpretation.
• Program, resident, and radiologist counts are stocks. Annual entries, graduates, retirements, and turnover are flows.
• The integrated IR sensitivity cannot be treated as a clean DR decade series.
• Medicare-enrolled radiologist counts vary by inclusion and activity rules across studies.

Appendix B. Full ACGME training series

Source reconstruction: ACGME Data Resource Books for 2018-19, 2021-22, and 2024-25. The integrated IR sensitivity begins in 2020-21 to avoid presenting early pathway counts whose comparability requires additional specialty-level validation. Blank combined values are intentional.

ACGME specialty update snapshot, 2025-26

The update listed three newly accredited diagnostic radiology programs: Mass General Brigham, University Medical Center of Southern Nevada, and Hackensack Meridian. New accreditation is a leading indicator of pipeline growth, but programs generally require phased recruitment before reaching a mature complement.

Appendix C. Proposed primary mixed-methods study

An explanatory sequential design with a national program census followed by purposive multiple-case interviews. Invite program directors, associate program directors, designated institutional officials, department chairs, GME finance leaders, and clinical operations leaders at all accredited diagnostic radiology programs, sampling integrated IR leadership as a prespecified linked stratum. The unit of quantitative analysis is the program-year. The unit of qualitative analysis is the teaching-site decision episode. The primary quantitative outcome is net change in funded resident complement over five years.

Quantitative hypotheses

1. Programs with higher protected core faculty FTE per resident will have greater odds of funded complement expansion.
2. Programs at hospitals above Medicare GME caps will be less likely to expand unless alternate institutional or state funding is available.
3. Recent accreditation citations related to faculty, supervision, evaluation, facilities, or learning environment will be associated with non-expansion or delayed activation.
4. Expansion tied to rural or underserved rotations will predict shortage-area practice only when paired with longitudinal recruitment and retention incentives.
5. Higher case-mix-adjusted workload and turnover will weaken the relationship between nominal faculty headcount and educational readiness.

Because program-level expansion is likely uncommon and predictors are correlated, the analysis should favor penalized logistic or ordinal regression and report uncertainty rather than rely on the conventional ten-events-per-variable heuristic. Nonresponse weighting should use observable ACGME program size, region, sponsor type, new-program status, and urban-rural affiliation.

Appendix E. Limitations, bias assessment, and robustness checks

Robustness checks completed

• All reported endpoint percent changes were independently recalculated.
• Training observations appearing in overlapping ACGME books were reconciled.
• Strict DR and combined DR plus integrated IR interpretations were both retained.
• UK and U.S. shortage measures were not directly compared.
• The apparent approved-minus-filled ACGME position difference was not labeled as a vacancy.
• Scenario overlap was presented as uncertainty, not as proof of balance.

References

Authoritative and peer-reviewed sources. Filter by category.

• Accreditation Council for Graduate Medical Education. (2019). Data Resource Book, Academic Year 2018-2019. Source
• Accreditation Council for Graduate Medical Education. (2022). Data Resource Book, Academic Year 2021-2022. Source
• Accreditation Council for Graduate Medical Education. (2025). Data Resource Book, Academic Year 2024-2025. Source
• Accreditation Council for Graduate Medical Education. (2026). Radiology Specialty Update 2025-2026. Source
• American College of Radiology. (2026, February 5). The radiologist shortage: A workforce update from HPI. ACR Bulletin. Source
• Begum, H., Short, H., Sandifer, S. P., Cottrill, R., Jordan, R. A., & Bruno, M. A. (2026). Potential to expand existing US residency programs in diagnostic radiology. Journal of the American College of Radiology, 23(5), 834-836. DOI · PubMed
• Centers for Medicare & Medicaid Services. (n.d.). Direct Graduate Medical Education and Section 126 residency slots. Source
• Christensen, E. W., Chung, Y. K., Rula, E. Y., & Parikh, J. R. (2024). Changes in the radiology practice landscape and indicators of practice consolidation from 2014 to 2023. American Journal of Roentgenology, 223(2), e2431357. DOI · PubMed
• Christensen, E. W., Liu, C. M., Rula, E. Y., & Parikh, J. R. (2026). Attrition of the national radiologist workforce: Associations with radiologist and practice characteristics. American Journal of Roentgenology, 226(1), e2533587. DOI · PubMed
• Christensen, E. W., Drake, A. R., Parikh, J. R., Rubin, E. M., & Rula, E. Y. (2025). Projected US imaging utilization, 2025 to 2055. Journal of the American College of Radiology, 22(2), 151-158. DOI · PubMed
• Christensen, E. W., Parikh, J. R., Drake, A. R., Rubin, E. M., & Rula, E. Y. (2025). Projected US radiologist supply, 2025 to 2055. Journal of the American College of Radiology, 22(2), 161-169. DOI · PubMed
• Government Accountability Office. (2021). Physician workforce: Caps on Medicare-funded graduate medical education at teaching hospitals (GAO-21-391). Source
• Government Accountability Office. (2025). Graduate medical education: Allocation of new Medicare-funded residency positions (GAO-26-107686). Source
• Health Resources and Services Administration. (2025). Physician workforce projections, 2023-2038. Source
• Jing, A. B., Garg, N., Zhang, J., & Brown, J. J. (2025). AI solutions to the radiology workforce shortage. npj Health Systems, 2(1), 20. DOI
• Malhotra, A., Kandala, H., Futela, S., Payabvash, S., Lakhani, P., Gandhi, D., Whitlow, C. T., & Duszak, R., Jr. (2026). The evolving United States radiologist pipeline. Journal of the American College of Radiology. DOI
• Parikh, J. R., Drake, A. R., Rula, E. Y., Golding, E., & Christensen, E. W. (2026). Radiologist turnover in the United States. Journal of the American College of Radiology, 23(6), 1058-1066. DOI · PubMed
• Rawson, J. V., Smetherman, D., & Rubin, E. (2024). Short-term strategies for augmenting the national radiologist workforce. American Journal of Roentgenology, 222(6), e2430920. DOI
• Rosenkrantz, A. B., & Cummings, K. W. (2025). Trends in emergency department imaging utilization among Medicare fee-for-service beneficiaries, 2013-2023. Radiology, 316(3). DOI
• Royal College of Radiologists. (2025). Clinical radiology workforce census 2025. Source
• Smith-Bindman, R., Kwan, M. L., Marlow, E. C., et al. (2019). Trends in use of medical imaging in US health care systems and in Ontario, Canada, 2000-2016. JAMA, 322(9), 843-856. DOI
• U.S. Census Bureau. (2016). Nation’s older population still growing. Source
• U.S. Census Bureau. (2017). Older Americans Month: May 2017. Source
• U.S. Census Bureau. (2019). By 2030, all baby boomers will be age 65 or older. Source
• U.S. Census Bureau. (2023). 2020 Census: 1 in 6 people in the United States were 65 and over. Source
• U.S. Census Bureau. (2025). Older adults outnumber children in 11 states and nearly half of U.S. counties. Source

Suggested citation

Emrick, K. (2026). The U.S. diagnostic capacity gap: Aging, imaging intensity, radiologist supply, and residency training, 2014-2055.

Disclosure and scope notes

This research is an independent secondary evidence synthesis. It does not represent the views of the cited agencies, journals, professional societies, or authors. No new human subjects were enrolled and no unpublished interview data are presented. The research is intended for health services research, workforce planning, and policy discussion, and is not clinical guidance. Search and verification cutoff: July 28, 2026. Projections are scenario-dependent and should be updated when source agencies revise data or methods.