AI Can Read an MRI in Seconds. Why are the Orders Still Locked in Unstructured Formats?

In the best-case scenario, ordering an MRI takes one click. A primary care physician decides a patient needs an MRI and enters the order into their electronic health record system. When both the doctor’s office and the imaging scanner belong to the exact same hospital network, that single click flows seamlessly into the central schedule, keeping patient records unified and avoiding manual delays.

But many referrals don’t stay inside one network. When a primary care doctor transmits an order across town to an independent outpatient imaging center, the request leaves the health system’s software behind. Because the imaging center operates as a separate business on a completely different system, the digital workflow halts. Much of the time, the order arrives as a static document. It can be a paper fax on a tray, a PDF in a shared inbox, or a printout with a cover sheet. Staff at the front desk must then manually read and retype every detail into their local computer, field by field, before the patient can even be scheduled.

eFax^®, a digital cloud fax and intelligent document processing provider, analyzed national cost and imaging-utilization data to map how much of the radiology referral process still runs on documents that staff must retype by hand. What the numbers show is a unique split: the technology that reads the scan is racing ahead, while the paperwork that requests the scan is, by one industry-wide measure, moving in the opposite direction.

Where radiology aimed its AI

More than 1,500 AI-enabled medical devices have now been authorized by federal regulators, and by independent tallies roughly three quarters of them were built for radiology: software that flags a bleed on a CT, measures a tumor on an MRI, reads a mammogram alongside the doctor. No specialty shows up on that list more, and all of that adoption landed on the machines that take the pictures.

The scans themselves are multiplying, too. Americans are projected to get 16.9% to 26.9% more imaging in 2055 than they did in 2023, driven mostly by an aging population that needs more of it.

The supply of doctors who read those scans is growing, too, just not fast enough. Demand is expected to grow faster over the same stretch, which is why the radiologist shortage is projected to persist for decades even as the profession adds people.

The workforce that does exist is churning, too. Practice turnover among radiologists rose 61% between 2013 and 2022, and the odds that a radiologist changed jobs in a given year were roughly twice as high from 2020 through 2022 as in 2013. The research ties that churn to workload: past a measurable threshold, the heavier a radiologist’s reading load, the likelier a move becomes. The squeeze compounds itself: past that point, rising demand raises workloads, and heavier workloads move more radiologists to change where they practice.

That squeeze puts a price on every wasted minute in the pipeline, and radiologists have long known where the slow minutes hide: at the front desk, where the requests come in.

Fifteen minutes here, five dollars there

Start with prior authorization. Before many scans, the patient’s insurer has to approve the request. When that approval is handled by hand, a staff member pulls the patient’s chart, fills out the insurer’s form, faxes it along with clinical notes, and then waits, often following up by phone when a decision doesn’t come back. Delays often come from more than slow transmission. Requests get sent back when clinical criteria are incomplete, such as unattached physical therapy logs or missing trial-of-therapy notes, and each round trip pushes back the date the patient can be scheduled. Modern clinical workflow software helps close this gap by scanning chart documentation and flagging missing payer requirements before the request is submitted.. All told, a manual prior authorization costs a provider about $12.95. Handled fully electronically, it costs about $4.71, a savings of $8.24 per request, and the electronic route gives staff back roughly 15 minutes each time.

Fifteen minutes sounds insignificant until it has a wage attached. The medical records specialists who do this kind of work earn a median wage of about $24.59 an hour. At that rate, every request that arrives as a fax or PDF, instead of as data the computer can read, costs roughly $6 in staff time just to retype.

Then there is the paperwork behind the paperwork. Insurers usually want proof that a scan is justified: the clinical notes, the exam findings, the order itself. In industry shorthand, those supporting documents are called attachments, and they are the closest thing industry transaction data has to a portrait of the referral paper trail.

Those attachments do not always travel as data a computer can use. The industry count behind these figures treats standardized transactions as fully electronic, while fax, email and mail count as manual channels even when the document itself is digital. In 2023, 29% of medical attachments moved electronically, and in 2024, that number increased to 32%. 

In 2025, that share fell to 24%. The regression was caused by the 2024 cyberattack on Change Healthcare, a major clearinghouse that many providers rely on to process claims. Doctors, hospitals, and pharmacies nationwide could not process bills, check insurance, or fill prescriptions smoothly. 

Since no federal standard for electronic attachments existed at the time, nothing required those connections to be rebuilt. A rule finalized in March 2026 sets the new standard for claims, with a 2028 compliance deadline. While these new standards resolve some issues around payments and prescriptions, they do not cover important document transmissions such as prior-authorization attachments. This leaves a list of important medical evaluation instruments, like scans and imaging diagnostics, exposed to further vulnerability in an already fragile system. 

The scale of what remains manual is the striking part. About 101 million of the roughly 142 million medical attachments required significant manual handling: faxed, mailed, or uploaded and retyped.

What would full electronic handling be worth? Across the medical side of the system, about $18.7 billion a year in savings. 

The quiet afternoon between the order and the callback

None of that math is visible from the patient’s side of the counter. What the patient knows is that the doctor ordered the scan this morning, the visit ended with a reassuring “they’ll call to schedule,” and now the phone is quiet.

The request is simply in line. Somewhere at an imaging center, it is waiting for a person to read it, check the coverage, confirm the permission slip, and key it into the schedule. Most of the time the call comes, the slot gets booked, and the scan happens.

The wait in between, the phone tag and the delayed care that follows,  the second recitation of an insurance card number: all of it is a document problem. And document problems are the fixable kind.

Why the request still rides a phone line

Inside a single health EMR system, this was solved years ago. A doctor can send an order to their own hospital’s imaging department without paper touching anyone. Electronic ordering inside the EMR of a single organization works.

The trouble starts at the property line. An independent imaging center is a separate business, running software the referring practice does not share and having patient identifiers such as medical record numbers that do not match those in other organizations. These organizations are not, by default, connected to each other; when they are, patient matching is not automatic. So when an order crosses that line, it drops to the one HIPAA-compliant, secure format any office can receive: a fax. Additionally, modern document processing now can also transform incoming documents directly into FHIR structured data that flows end-to-end-into the receiving system.

Organizations have a keen interest in ensuring these referral and prior authorization faxes are responded to quickly. In addition to patient satisfaction and revenue concerns, there are also regulatory considerations to keep in mind. Since January 2026, major government-regulated insurers face time limits on authorization decisions, and by January 2027 they must offer fully electronic prior authorization through standard connections.

The reading room was first

None of these are new processes; what is changing is the expanse of how the technology can help. The first wave of imaging AI went to the reading room. A newer category, Intelligent Document Processing, or AI extraction, points at the intake desk instead: software that reads an incoming fax or PDF the way a staff member would, classifies it, understands the level of urgency associated with it, and then extracts relevant data into the applicable systems for scheduling and approval. Health systems are beginning to pair that extraction layer with the cloud fax lines they already run. Software vendors are building it into intake workflows. In many cases, the same technology can also support new FHIR translation regulations. Federal rule-makers, on a fixed deadline, are pushing the insurance half of the transaction onto electronic rails. None of it is finished, and none of it requires either side of the property line to replace its systems. The fix is aimed at the documents themselves.

A machine can image a spine in minutes, and software can now support the radiologist in reading the result in seconds. The slow part is often the front-end request processes: requesting the referral, getting the authorization to do the imaging, exchanging the document communications to get that authorization. The papers or digital communications only move as fast as the clinic day allows.

That distance is starting to close as intelligent document processing adds relief to clinical days, ensuring a reduction in the time it takes to support front-end tasks that frequently delay patient care.