Medical imaging

The Cost of Seeing Inside the Body

A first-principles account of how five imaging systems turn physics into decisions, why the entry cost is falling, and which constraints do not yield to software.

Research through 19 September 2026
Figure 1

How far the entry floor has fallen

The curve estimates the purchase price of an entry-level clinically useful system, expressed in constant 2026 US dollars. It is a reconstruction from sparse published benchmarks, not a market price index.

Entry purchase cost of five medical imaging modalities from 1950 to 2025A logarithmic chart in constant 2026 US dollars. Ultrasound shows the largest decline. CT, MRI, and PET begin when commercial systems appeared.$10m$1m$100k$10k$1k19501960197019801990200020102020X-rayUltrasoundCTMRIPET
Reading the curve

Select a modality to isolate its reconstructed entry-cost path. Connected lines interpolate between sparse published benchmarks; they are not annual market observations.

Figure 1: Representative entry-system purchase price, 1950 to 2025, on a logarithmic scale. Connected segments are interpolation. Product capability, configuration, geography, installation, and service differ across observations. X-ray is included from 1950; each later modality begins near commercial introduction. Premium 2025 systems can sit far above the plotted entry boundary. Historical nominal prices were approximately converted with US CPI, then rounded aggressively to avoid false precision.
80xApproximate fall from an early room-scale ultrasound benchmark to today's probe-class entry floor.
30x+Approximate gap between early whole-body MRI in today's money and the low end of portable low-field systems.
Not one marketA $3,000 probe and a premium radiology platform answer different questions. Falling entry cost does not imply equal diagnostic coverage.
Read the assumptions and benchmark record

1960s ultrasound: cross-sectional systems up to $24,000 nominal; simpler measurement units were far cheaper.

1978 CT: the US GAO reported $300,000 to $700,000 per scanner, plus several hundred thousand dollars annually for operation and maintenance.

1980s MRI: published accounts put systems near $1 million nominal, with installation and annual support adding substantially more.

1996 PET: a US Veterans Affairs review cited $800,000 to $2.5 million, excluding construction and the cyclotron.

2025 endpoints: low-field MRI estimates and a documented 2024 PET/CT quote establish lower market boundaries, not median prices.

What the record shows

  • The cheapest useful scanner is improving faster than the most capable scanner is getting cheap. Handheld ultrasound and portable low-field MRI open new settings, while premium CT, MRI, and PET remain capital-intensive.
  • CT moved anatomy from overlapping projections to volumes. Photon-counting detectors now improve resolution and energy discrimination while offering routes to lower dose or contrast burden.
  • Computation increasingly substitutes for acquisition time and signal. That makes training data, uncertainty, failure detection, and reconstruction provenance part of the medical device.
  • The full cost is not the invoice. Construction, uptime, staffing, throughput, interpretation, repeat scans, and the cost of a missed diagnosis define the delivered economics.

This report separates equipment purchase price, facility cost, per-exam production cost, hospital charge, reimbursement, and patient payment. They answer different questions and should not be placed on one curve.

Figure 2: The imaging trade space

Every image spends a different budget

Time, dose, field strength, computation, infrastructure, and operator skill trade against one another.

High soft-tissue contrast and many functional contrasts

Signal
Magnetic field and radiofrequency response
Acquisition
Minutes
Footprint
Shielded suite or portable low-field unit
Ionizing radiation
No

Binding constraint: Cost, scan duration, magnetic safety, and access

“Better” depends on the clinical question. Portability can increase access while reducing resolution; higher fidelity can increase facility cost, scan time, or dose.
Part I: First principles

An image is an estimate under constraints

The body does not emit a ready-made picture. A scanner perturbs tissue, records a noisy response, and reconstructs an estimate that must be good enough for one decision.

Clinical valuedecision improvement × access × timeliness÷dose + errors + total delivered cost
01

Create contrast

An image starts with a physical difference between tissues: attenuation, acoustic impedance, proton relaxation, or tracer uptake. If disease does not change the chosen signal, processing cannot reveal it.

Measure
Contrast-to-noise, tissue specificity
Failure boundary
Weak or non-specific biological contrast
What the record shows at this step
Reported evidence
Iodine and gadolinium agents, and targeted radiotracers, exist precisely because native contrast is often insufficient for the clinical question.
Where it is moving
Spectral and photon-counting acquisition separates materials by energy, and molecular tracers extend contrast to receptor and metabolic states.
Principal risk
Weak or non-specific biological contrast

Evidence statements describe cleared systems or published studies. “Where it is moving” is an editorial reading of the direction, not an announced capability.

02

Deliver energy or tracer

X-rays, sound, radiofrequency pulses, magnetic gradients, or radiotracers probe the body. More signal can improve certainty, but may add dose, heating, time, or biological burden.

Measure
Dose, field strength, acoustic output
Failure boundary
Radiation, heating, contrast, and tracer exposure
What the record shows at this step
Reported evidence
The FDA publishes a broad typical effective-dose range of roughly 1 to 10 mSv for diagnostic CT, with wide variation by exam and protocol.
Where it is moving
Total-body PET converts a large sensitivity gain into either a much shorter scan or a substantially lower administered dose.
Principal risk
Radiation, heating, contrast, and tracer exposure

Evidence statements describe cleared systems or published studies. “Where it is moving” is an editorial reading of the direction, not an announced capability.

03

Detect the return

Detectors and coils turn small physical responses into electrical measurements. Higher efficiency can buy lower dose, shorter scans, or finer resolution.

Measure
Quantum efficiency, bandwidth, dynamic range
Failure boundary
Electronic noise, drift, saturation, and hardware cost
What the record shows at this step
Reported evidence
Digital radiography removed film entirely; the first photon-counting CT system was cleared in 2021 and records individual photons and their energy.
Where it is moving
Semiconductor-on-chip ultrasound transducers and denser PET crystal arrays move detector cost onto a semiconductor cost curve.
Principal risk
Electronic noise, drift, saturation, and hardware cost

Evidence statements describe cleared systems or published studies. “Where it is moving” is an editorial reading of the direction, not an announced capability.

04

Sample enough evidence

A system records one projection or many views through space and time. Motion and anatomy determine how much data is enough for the clinical question.

Measure
Coverage, temporal resolution, acquisition time
Failure boundary
Motion, incomplete angles, patient throughput
What the record shows at this step
Reported evidence
Undersampled MRI acquisition with constrained reconstruction is commercially deployed and shortens examinations at comparable reported quality.
Where it is moving
Acquisition that stops when the evidence is sufficient, rather than when a fixed protocol ends, is the open research direction.
Principal risk
Motion, incomplete angles, patient throughput

Evidence statements describe cleared systems or published studies. “Where it is moving” is an editorial reading of the direction, not an announced capability.

05

Reconstruct an estimate

Algorithms convert noisy, incomplete measurements into an image. Iterative and learned methods exchange computation and prior assumptions for time or dose.

Measure
Latency, artifact rate, uncertainty
Failure boundary
Suppressed pathology or invented structure
What the record shows at this step
Reported evidence
Learned reconstruction is cleared and in routine use on major CT and MRI platforms, where it substitutes computation for acquired signal.
Where it is moving
Calibrated uncertainty and failure detection, so a reconstruction can report when its prior is doing more work than the measurement.
Principal risk
Suppressed pathology or invented structure

Evidence statements describe cleared systems or published studies. “Where it is moving” is an editorial reading of the direction, not an announced capability.

06

Make a decision

Clinicians combine the image with history, priors, laboratory results, and consequences of error. The valuable output is a better decision, not a prettier picture.

Measure
Time to report, sensitivity, specificity
Failure boundary
Incidental findings, automation bias, limited workforce
What the record shows at this step
Reported evidence
A 2026 meta-analysis found moderate performance for portable low-field MRI in acute ischemic stroke and cautioned against using it alone to exclude disease.
Where it is moving
Evidence that faster or nearer imaging changes outcomes, rather than only changing where and how often images are made.
Principal risk
Incidental findings, automation bias, limited workforce

Evidence statements describe cleared systems or published studies. “Where it is moving” is an editorial reading of the direction, not an announced capability.

Resolution is not fidelity. Fidelity is preserving the evidence needed for the decision while exposing the uncertainty.

Five modalities, five kinds of truth

No scanner sees everything. Each modality makes a different tissue property visible and inherits a different physical limit.

ModalityPrimary contrastStrongest useAccess advantageBinding trade-off
X-rayDifferential attenuationBone, chest, lines and tubesVery fast, mobile, inexpensiveProjection overlap, ionizing dose
UltrasoundReflected sound and Doppler shiftMotion, pregnancy, heart, vessels, proceduresReal time, handheld, no ionizing doseOperator skill, bone and air block sound
CTMany X-ray projectionsTrauma, lung, vessels, acute abdomenFast 3D anatomy, broad availabilityDose, contrast, fixed infrastructure
MRINuclear spin responseBrain, spine, joints, soft tissue, functionMany contrasts, no ionizing doseTime, siting, safety, capital burden
PETRadiotracer annihilation photonsMetabolism, receptors, treatment responseMolecular sensitivityTracer logistics, dose, coarse spatial detail

Who is building what

Detector physics, reconstruction software, portability, and price are being pushed by different organisations with different claims. Search the record, or filter by the layer each programme is working on.

10 programmes
Siemens HealthineersNaeotom AlphaPhoton-counting detector CT
Reported evidence
Cleared by the FDA in 2021 as the first photon-counting clinical CT system; published studies report higher spatial resolution and intrinsic spectral data.
Announced next step
Wider deployment of photon-counting platforms across routine and lower-cost configurations.
Unresolved risk
Premium pricing, detector supply, and evidence that resolution and spectral data change management rather than only image appearance.
GE HealthCareRevolution / AIR Recon DLDeep-learning reconstruction across CT and MRI
Reported evidence
Cleared learned reconstruction is in routine use and is reported to shorten MRI acquisitions at comparable diagnostic quality.
Announced next step
Broader protocol coverage and shorter standard examinations.
Unresolved risk
Generalisation across scanners, body types, implants, and rare pathology; learned priors can suppress or invent structure.
PhilipsAmbition / Compressed SENSESealed low-helium magnets and undersampled acquisition
Reported evidence
Commercial systems reduce helium dependence and cut scan time through undersampling with constrained reconstruction.
Announced next step
Lower siting and service burden for conventional field strengths.
Unresolved risk
Capital cost remains high; siting and shielding still dominate the installed price.
HyperfineSwoopPortable ultra-low-field (0.064 T) brain MRI
Reported evidence
FDA-cleared portable system studied at the bedside; a 2026 meta-analysis found moderate performance in acute ischemic stroke.
Announced next step
Bedside neuroimaging in intensive care, emergency, and resource-limited settings.
Unresolved risk
Sensitivity is not equivalent to high-field MRI; the published caution is against using it alone to rule disease out.
Butterfly NetworkiQ3Semiconductor-on-chip whole-body probe
Reported evidence
A single probe covering multiple exam types at a probe-class price point, sold into point-of-care settings.
Announced next step
Guided acquisition that lowers the operator-skill barrier.
Unresolved risk
Operator dependence, scan quality assurance, documentation, and the risk of low-value imaging outside a referral pathway.
United ImaginguEXPLORERTotal-body PET with ~194 cm axial coverage
Reported evidence
Reported order-of-magnitude sensitivity gains, enabling much shorter scans or substantially reduced tracer dose.
Announced next step
Lower dose per study and higher patient throughput per scanner.
Unresolved risk
Very high capital cost and a small installed base; economics depend on utilisation and reimbursement.
Nano-X ImagingNanox.ARCCold-cathode source tomosynthesis
Reported evidence
A cleared system built around a different source technology, positioned on cost rather than capability.
Announced next step
Lower-cost distributed screening and musculoskeletal imaging.
Unresolved risk
Clinical adoption, throughput, service networks, and evidence of real diagnostic yield remain unsettled.
Refurbished equipment marketMulti-vendorCertified second-life CT, MRI, and PET/CT systems
Reported evidence
Documented purchase records show refurbished and entry systems well below premium list prices.
Announced next step
Extending equipment life and lowering the entry boundary for new sites.
Unresolved risk
Parts support, cybersecurity patching, and software licensing can end useful life before the hardware fails.
NIH / NIBIBPublic research programmesDetector physics, reconstruction, and access research
Reported evidence
Sustained public funding underlies detector, tracer, and reconstruction advances later commercialised.
Announced next step
Methods that lower dose, cost, or time without lowering diagnostic yield.
Unresolved risk
Laboratory performance does not establish clinical benefit; translation timelines are long.
FDA device programmesRegulatory pathwayClearance and post-market surveillance of imaging devices
Reported evidence
Clearance summaries define intended use, and dose guidance publishes typical CT effective-dose ranges.
Announced next step
Frameworks for adaptive and learned components in cleared devices.
Unresolved risk
Clearance is not evidence of outcome benefit, and locked-model clearance sits awkwardly with continuously updated software.

Systems and status are reproduced from the research record below. A clearance is a statement about intended use and safety, not about outcome benefit, and naming a system here does not imply an endorsement or a procurement recommendation.

The access frontier is moving

0.064 TPortable ultra-low-field MRI has been studied for bedside brain imaging.
2021The first photon-counting CT system received FDA clearance.
1-10 mSvThe FDA's broad typical effective-dose range for diagnostic CT, with large exam and protocol variation.

Portability is not equivalence. A 2026 meta-analysis found moderate performance for portable low-field MRI in acute ischemic stroke and cautioned against using it alone to rule disease out.

The discovery chain

Imaging advances when a new signal, a better detector, and enough computation become clinically usable at the same time.

  1. 1895

    X-rays

    A new penetrating signal made internal anatomy visible without surgery.

  2. 1950s

    Clinical ultrasound

    Pulse-echo systems adapted sonar physics to living tissue.

  3. 1971

    First clinical CT

    Reconstruction separated overlapping anatomy into cross-sections.

  4. 1980s

    Commercial MRI and PET

    Magnetic resonance expanded soft-tissue contrast while PET mapped molecular activity.

  5. 1998-2001

    PET/CT

    Registration of molecular and anatomical images moved from prototype to commercial systems.

  6. 2021

    Photon-counting CT

    The first FDA-cleared system counted individual X-ray photons and their energy.

  7. 2020s

    Portable MRI and AI reconstruction

    Lower-field magnets and computation began moving advanced imaging toward the bedside.

Part II: The cost-down engine

More answer per unit of infrastructure

The durable trend is not that every flagship machine gets cheaper. It is that useful information migrates into smaller systems, shorter protocols, and wider clinical settings.

01

Detectors count more of the signal

Digital radiography removed film. Solid-state ultrasound and PET detectors improved sensitivity. Photon-counting CT extracts energy information from individual events.

Cost lever: less dose, fewer repeats, more information per acquisition
02

Computation replaces brute force

Iterative reconstruction, compressed sensing, motion correction, and learned denoising can recover useful images from faster or weaker acquisitions.

Cost lever: shorter slots and lower infrastructure demand
03

Hardware leaves the suite

Mobile radiography, handheld ultrasound, and portable low-field MRI move the scanner to emergency rooms, intensive care, rural clinics, and bedsides.

Cost lever: less patient transport and lighter siting
04

Protocols become task-specific

A focused ultrasound or rapid MRI protocol can answer one narrow question without reproducing a comprehensive radiology examination.

Cost lever: match evidence to the decision
05

Networks pool scarce expertise

Cloud archives, teleradiology, remote quality control, and decision support separate image acquisition from specialist location.

Cost lever: higher utilization and wider interpretation access
06

Manufacturing gains a second market

Refurbishment, modular platforms, and software upgrades extend equipment life, though cybersecurity and parts support can end it early.

Cost lever: spread embodied capital across more years
Part III: The delivered price

The invoice is a small part of the cost

A scanner has one price. An examination has a production cost that depends on the room, the service contract, the staff, the tracer, and above all on how often the machine is used.

The cost stack

Equipment and finance
Room, power, shielding, cooling
Service and downtime
Operator and interpretation
Contrast, tracer, disposables
Low utilization and repeat scans

Bars are qualitative, not shares. Their purpose is to show why a cheaper magnet or detector does not translate one-for-one into a cheaper examination. The model below makes the same argument quantitatively.

Figure 3 · Interactive model

What does one examination cost to produce?

Production cost is not a charge, a reimbursement, or a patient bill. Change the assumptions and watch which line dominates.

Community CT$118/exam

Estimated production cost

Capital recovery
$17 · 15%
Service and maintenance
$10 · 9%
Operator and interpretation
$55 · 47%
Contrast, tracer, disposables
$35 · 30%

10-year equipment life; 8% real cost of capital; annual service 9% of installed cost; 50 operating weeks. Excludes construction financing, referral and scheduling overhead, repeat scans, and the cost of a missed or false-positive finding.

Calculation & assumptions

Cost per exam = installed cost × (capital recovery factor + annual service rate) ÷ annual examinations + per-exam labour + per-exam consumables. Installed cost = equipment price × (1 + siting percentage).

Capital recovery uses the scenario's real discount rate and equipment life. Utilisation is expressed directly as examinations per week, which is where low-volume sites lose most of their economics: halving throughput roughly doubles the capital and service share per exam.

An editorial model, not a price index. Its purpose is to show why a cheaper scanner does not produce a proportionally cheaper examination: for ultrasound the dominant line is labour, for MRI it is capital, and for PET it is the tracer.
UtilisationCapital and service are fixed annual costs divided by throughput. A half-used scanner carries roughly twice the capital cost per examination.
LabourFor the cheapest hardware, the operator and the report are the cost. A $3,000 probe does not produce a $3 examination.
ConsumablesA PET tracer can exceed every other line combined, which is why sensitivity gains that cut dose also cut cost.
Part III: Where progress is stuck

The remaining bottlenecks are system bottlenecks

Imaging can fail even when the scanner works perfectly. Physics, workflow, institutions, and evidence all set the boundary.

The signal budget

Spatial resolution, time, dose, penetration, and noise are coupled. Better algorithms stretch the budget; they do not repeal counting statistics or tissue physics.

The room around the scanner

Power, shielding, cooling, coils, injectors, service, safety zones, and construction can rival the machine in determining access.

People and workflow

A device needs trained operators, protocol selection, quality assurance, reporting, follow-up, and a referral system. Hardware alone does not produce care.

Evidence and generalization

A reconstruction model may fail on an unfamiliar scanner, body type, implant, disease, or protocol. Prospective clinical validation is slower than software iteration.

The last-mile price

Purchase prices can fall while patient prices remain high because utilization, reimbursement, staffing, financing, and local market power set the delivered cost.

Overuse and incidental findings

More accessible imaging can create low-value scans and cascades of follow-up. The frontier is the right scan at the right time, not simply more scans.

An optimistic view, with conditions

Imaging becomes a distributed measurement layer

The strongest plausible future is not one universal scanner. It is a ladder: ubiquitous low-cost tools identify who needs escalation, fast local systems answer common questions, and high-end centers resolve the difficult remainder.

Now to 2030

Focused exams spread

Handheld ultrasound, rapid protocols, photon-counting CT, and portable MRI expand where imaging can happen. Workflow and training determine the real gain.

2030s

Acquisition gets adaptive

Systems stop when sufficient evidence is collected, choose views from live uncertainty, and route equivocal cases upward instead of pretending every image is definitive.

Longer horizon

Price follows the whole pathway

Payment rewards avoided delays and better outcomes, not scan volume alone. Shared infrastructure and remote expertise turn falling hardware cost into falling delivered cost.

View the analyst probability ranges
MilestoneDateAnalyst probability
Photon-counting CT available at most large referral centres203250-70%
Portable low-field MRI routine in stroke and intensive-care pathways203230-50%
Learned reconstruction in the default protocol on most new scanners203060-80%
A large trial shows portable imaging improves outcomes, not only access203320-35%
Delivered price of a common cross-sectional exam falls materially in a high-cost market203515-30%
Autonomous reporting cleared for a narrow, defined screening task203225-45%

Editorial judgements conditional on the record above. They are not vendor guidance, published forecasts, or statistical confidence intervals.

There is more than one finish line

  1. Technically readableThe system produces a recognizable image.
  2. Diagnostically adequateThe evidence answers one defined clinical question.
  3. Workflow compatibleAcquisition, interpretation, and reporting fit real care.
  4. Economically accessibleTotal delivered cost supports broad and equitable use.
  5. Outcome improvingEarlier or better decisions materially improve health.

Sources, method, and boundaries

The historical figure is an editorial reconstruction, not a continuous price series. It combines documented purchase-price anchors with rounded CPI conversion and approximate entry-market endpoints. It does not compare equal capability and should not be used for procurement or patient billing. Clinical performance statements describe populations and systems in cited research, not advice for an individual patient.

Measured
A reported physical or clinical observation.
Purchase price
Equipment cost before some or all construction, service, staffing, and consumables.
Entry boundary
A lower available tier that may have narrower indications than a premium system.
Analytical reconstruction
A connected estimate from sparse, non-like-for-like benchmarks.