Deep Learning MRI Reconstruction Cost for OEM Upgrades, SwiftMR and SubtleHD

The short answer: deep learning MRI reconstruction from the scanner maker costs about $145,000 to $150,000 per scanner on the most recent federal purchase orders: $144,574 for Siemens Deep Resolve on a MAGNETOM Sola and $149,210 for Philips SmartSpeed, both bought by the VA. Vendor-neutral software such as AIRS Medical SwiftMR and Subtle Medical SubtleHD has no published price and no federal purchase on record. It is quoted as a subscription per scanner or per study.

This page is for the person who has to sign the quote: the owner or COO of an outpatient imaging center, or the imaging director at a hospital, who has been told that deep learning reconstruction will add MRI slots without a new magnet and now has to decide which version to buy. Every dollar figure below is from a named federal purchase order you can look up yourself. Where the record is empty, we say so instead of guessing.

Purchase data comes from USAspending.gov, award types A to D, searched by keyword on October 15, 2026, with every recipient name and description read by hand. Clearance data comes from the openFDA 510(k) database the same day.

How much does deep learning MRI reconstruction cost?

For the scanner makers' own products, the public record gives a fairly tight range. These are the federal orders whose descriptions name the product.

Two cautions about reading this table. First, Canon uses the AiCE name on both its MRI and CT scanners, and neither Canon order says which one it covered, so treat those two rows as a lower bound rather than an MRI price. Second, a federal order is one buyer's negotiated price on one date. It is a benchmark for your own quote, not a list price.

GE HealthCare's AIR Recon DL has three FDA clearances under that name (K213717 in 2022, K252379 in December 2025 and K253808 in August 2026), but we found no federal order that names it in the description, which can happen when a product is bought as part of a broader software level or scanner refresh. That is the next point.

Why one customer was quoted half a million dollars

In an AIRS Medical press release, the COO of First Choice Imaging in Utah said getting GE or Siemens deep learning on their scanners would have needed a system upgrade costing upwards of half a million dollars. That figure and the VA's $144,574 are not in conflict. They describe different starting points.

The VA Dallas orders added Deep Resolve to a Sola, a current Siemens platform that already supports it. An older magnet may need a new software level, new computing hardware or new coils before the maker will enable deep learning reconstruction at all, and that is where the quote climbs. So the first question to put to your OEM rep is whether each of your scanners qualifies as it stands, and what it would take for the ones that do not. Ask for that answer per scanner, in writing.

How much does SwiftMR cost?

AIRS Medical does not publish a price, and USAspending shows no federal award naming SwiftMR or AIRS Medical. A trap if you search this yourself: the keyword "AIRS MEDICAL" returns air ambulance and air medical transport contracts, none of which have anything to do with MRI.

What AIRS Medical does publish is its own return figures: on average a 38% increase in patient throughput, $44,000 a month in added revenue and an $8,000 monthly cut in operating costs at its customer sites. Those are vendor numbers, not an independent study, and they are averages across sites with very different case mixes. Use them to frame the conversation, then run a pilot on your own protocols before you rely on them.

SwiftMR itself has five FDA clearances, from K210999 in October 2021 to K261349 in July 2026, and AIRS Medical has announced a clearance that lets it run alongside OEM deep learning reconstruction rather than instead of it. That matters for a fleet where one scanner already has the OEM option and the others do not.

How much does Subtle Medical cost?

Subtle Medical does not publish a price either, and there is no federal award naming Subtle Medical, SubtleMR or SubtleHD. The product line has moved on since most comparison articles were written: SubtleMR (cleared in 2019, 2021 and 2023) has been succeeded by SubtleHD, cleared as K243250 in February 2025 and sold in the SubtleELITE suite with SubtleSYNTH and SubtleALIGN. In May 2026 the company added SubtleHD-PET and SubtleHD-CT. If you are quoted SubtleMR today, ask whether the price covers SubtleHD and which clearance applies.

Both vendor-neutral products are usually quoted as a recurring subscription, per scanner or per study. That changes the arithmetic, because a one-time OEM upgrade is a capital purchase and a subscription runs for as long as you use it.

OEM upgrade or vendor-neutral subscription, which costs less?

It depends on two numbers you can measure before signing: how many scanners qualify for the OEM option without extra hardware, and how many added exams a day your technologists actually achieve with shorter protocols.

A worked example, using only the public figure and an assumption you should replace with your own pilot data. Take the $144,574 Deep Resolve order. If shorter protocols add two exams a day on that scanner, over 260 working days that is 520 added exams a year, or 2,600 over five years. Spread across five years, the upgrade costs about $56 per added exam. If the real gain is one exam a day, it is about $111. A vendor-neutral subscription priced per study can be compared on exactly the same basis, and so can a per-scanner subscription divided by the added exams it produces.

Vendor-neutral subscriptions billed per study also tend to grow quietly as volume rises, and they sit outside the capital budget where most imaging centers watch MRI spending. If you run several per-study software subscriptions alongside PACS and AI tools, keep every recurring software subscription and its usage charges in one view, so the bill does not drift upward unnoticed.

What to put in the quote before you sign

• The K number for your use. Subtle Medical and AIRS Medical have both moved products from product code LLZ to QIH, and the OEM versions are cleared inside scanner submissions under LNH. Ask which clearance covers the sequences and body regions you plan to shorten.
• Eligibility per scanner. For OEM options, a written list of which of your scanners qualify as they stand and the cost to bring the others up.
• The unit of price. Per scanner, per study or one-time. For per-study pricing, the price at your projected volume in year three, not today.
• Pilot terms. A paid or free pilot on your own protocols, with the throughput gain measured by your team, before the full term starts.
• Renewal and exit. The renewal uplift cap, and what happens to the protocols you built around the product if you stop paying.

What the extra slots do to your reading room

Every added MRI slot is another study for the same radiologists to read. A center that gains two slots a day on each of three magnets adds about 1,500 MRI reads a year without adding a radiologist, and turnaround is usually where that shows up first. Faster scanning moves the bottleneck from the magnet to the reading room.

That is the part Radiological.ai covers. It reads from your PACS over standard DICOM, whatever made the scanner or the acceleration software, flags suspected findings, puts urgent studies at the top of the worklist and drafts the structured report for the radiologist to review, edit and sign. MRI is included on the Group plan at $598 per radiologist a month, or $299 a month billed annually. The comparison of every acceleration option, with each vendor's full FDA record, is on our Subtle Medical alternatives and SwiftMR comparison, and radiology worklist prioritization software shows how urgent studies stay first as volume climbs. For the read itself, see AI MRI analysis, and for what other radiology software costs, the radiology software pricing benchmarks.

Frequently asked questions

Is deep learning MRI reconstruction worth it for an imaging center?

It is worth it when scanner time is the bottleneck and you can staff and read the extra exams. At the VA's price of about $145,000 for Deep Resolve, two added exams a day work out to roughly $56 per added exam over five years. If patients are not waiting for MRI slots, shorter protocols mostly buy idle time.

Can SwiftMR work with GE AIR Recon DL or Siemens Deep Resolve?

AIRS Medical has announced an FDA clearance for SwiftMR to work in conjunction with OEM deep learning reconstruction. In practice that lets a center keep the OEM option on a newer scanner and use SwiftMR on the rest of the fleet. Ask AIRS Medical which K number covers the combination on your scanner models.

Does deep learning reconstruction require a new MRI scanner?

Not for vendor-neutral software, which works on images after the scanner has reconstructed them. OEM products such as AIR Recon DL, Deep Resolve, SmartSpeed and AiCE need a supported model and software level, and older scanners may need hardware or a software upgrade first, which is why some quotes run far above the VA's $145,000.