AI is already changing the work of allied health professionals, including radiographers, physiotherapists and laboratory scientists, often faster than it is changing doctors’ work. Tools for radiotherapy contouring, digital physiotherapy and laboratory screening are in routine use. Yet workforce planning, training and regulation rarely address these professions, which is where the least discussed shifts are happening.
Ask a health system leader how AI will affect the workforce and the answer almost always concerns doctors: will it replace radiologists, reduce burnout, change training? Occasionally nurses come up. Almost nobody mentions the radiographer who positions the patient and runs the scanner, the therapeutic radiographer who outlines organs on a planning scan, the physiotherapist with a waiting list of back pain referrals, or the biomedical scientist screening slides in a laboratory.
These allied health professionals make up a large share of the clinical workforce in most health systems. Their work is often more structured, more protocol-driven and more repetitive than a physician’s, which is exactly the kind of work AI handles well. In several areas, AI has already moved into their daily routine with less attention than any physician-facing tool.
The thesis here is that allied health is where AI’s workforce effects are arriving first and being planned for least. That creates two risks: opportunities to relieve severe shortages are missed, and changes to roles happen by default, through procurement decisions, rather than by design.
How is AI used by radiographers?
Radiography shows both sides of the shift.
In radiotherapy, one of the most time-consuming tasks is contouring: outlining the tumor and nearby healthy organs on planning scans so treatment can target the cancer and spare everything else. It can take a skilled professional a long time per patient. In September 2023, England’s National Institute for Health and Care Excellence published an early value assessment recommending that AI technologies could be used to help contour organs at risk for radiotherapy planning while further evidence is gathered. It was explicit that every AI-generated contour must be reviewed and edited as needed by a trained healthcare professional.
NICE’s analysis also made a telling point about economics. With a senior radiographer doing the contouring, the most expensive AI technology, at about £50 per plan, would need to save around 47 minutes to be cost neutral, while the cheapest, at about £4, would need to save only around 4 minutes. In other words, the value of the tool depends on whose time it saves and what that time is worth. That is a workforce question, not a technical one.
In diagnostic imaging, AI is also changing the radiographer’s day through automated positioning checks, image quality assessment, dose optimization and faster reconstruction for MRI and CT. These tools rarely make headlines, but they change throughput and what radiographers spend attention on.
The staffing context makes this urgent. The Society of Radiographers’ 2022 census of the UK diagnostic radiography workforce reported an average vacancy rate of 12.8% in England. When the same imaging departments are short of both radiologists and radiographers, AI that helps only the radiologist can simply move the bottleneck to the scanner.
Can AI replace a physiotherapist?
For some patients with some conditions, AI-driven services are already delivering physiotherapy programs without a human physiotherapist in each session.
Flok Health, a UK company, operates what it describes as the first digital musculoskeletal clinic approved by the Care Quality Commission, England’s health and care regulator. Patients with back pain have an automated assessment via smartphone video, and those judged suitable receive weekly sessions with an AI physiotherapist built from video of human clinicians. In January 2025, National Health Executive reported that the service was being offered to patients in NHS Lothian in Scotland, and that in trials all patients surveyed rated the experience as the same as or better than what they would expect from a human physiotherapist, with 86% reporting symptom improvement.
That is a meaningful development, particularly for musculoskeletal services with long waiting lists. It is also narrow. The service is for suitable patients with specific conditions, complex cases are routed to clinicians, and patient-reported satisfaction from company-linked trials is not the same as independent comparative outcomes.
The more realistic picture is not AI replacing physiotherapists but AI absorbing a share of routine, protocol-driven rehabilitation, while human physiotherapists concentrate on complex, post-surgical, neurological and frail patients. That could be a good use of a scarce workforce. It could also leave the human caseload heavier and more complex without anyone planning for it.
How is AI changing laboratory work?
Laboratories were automating long before the current AI wave. Analyzers, sample tracking and robotic handling have transformed clinical chemistry and hematology over decades. The newer shift is in morphology and image-based work, including pathology slides, cytology and blood films, where AI can pre-screen, prioritize and flag.
In 2021 the US Food and Drug Administration authorized Paige Prostate, the first AI software for identifying areas suspicious for cancer on digitized prostate biopsy slides. Many more digital pathology and laboratory tools have followed.
For laboratory scientists, the effect is similar to radiography. AI moves the scientist’s attention from routine screening toward reviewing flagged cases, quality assurance and managing digital infrastructure. The skills needed shift from long hours at the microscope toward validation, data management and exception handling.
There is a training consequence too. Much of a laboratory scientist’s pattern recognition was built by screening large volumes of mostly normal material. If AI takes the normal cases and leaves only the flagged ones, new staff see a skewed sample from their first week, and the judgment that came from volume has to be built some other way, through deliberate teaching sets and periodic unassisted review.
Laboratory work is also where the invisibility problem is sharpest. Patients never meet the biomedical scientist. Workforce shortages in laboratories rarely make the news, and changes to their work rarely feature in AI strategies.
A worked example: one radiotherapy department
Consider a hypothetical radiotherapy department planning 1,500 treatment courses a year, where contouring organs at risk averages 90 minutes of therapeutic radiographer or dosimetrist time per plan, and the team carries two vacancies.
The department adopts AI autocontouring. After local validation, the AI produces acceptable draft contours for most plans, and editing takes around 30 minutes on average. That saves about an hour per plan, roughly 1,500 hours a year, close to the annual working hours of one full-time professional.
Three choices follow. The department can use the time to reduce waiting times for the start of radiotherapy, which matters for outcomes in several cancers. It can redeploy staff into adaptive radiotherapy, where plans are adjusted during treatment, which needs more contouring, not less. Or it can stop recruiting to one of the vacancies.
None of these choices is inherently wrong, but they have different implications for patients and staff. And the hour saved depends on things the department controls: training staff to edit AI contours efficiently and safely, auditing contour quality, and keeping the manual skill for cases where the AI performs poorly, such as unusual anatomy or post-surgical changes.
If nobody in workforce planning is looking at this department, the choice gets made by default, usually as the vacancy that is quietly never filled.
What is the strongest counterargument?
The strongest objection is that allied health professions are protected by their hands-on nature. Radiographers position patients and operate equipment; physiotherapists touch, mobilize and treat; laboratory scientists handle samples and troubleshoot instruments. On this view, AI will support these professions at the margins and workforce concerns are overblown.
There is truth in that. Much allied health work is physical and relational, and it will remain so. But the objection misses where AI is actually landing: the cognitive, protocol-driven tasks around the physical work, such as contouring, image quality checks, exercise prescription and slide screening. Those tasks make up a substantial share of working time in several professions. Changing them changes the job, the training needed and the staffing model, even if the hands-on core remains.
The counterargument also underestimates speed. Allied health tools often face fewer professional and political barriers than physician-facing AI. They are bought by departmental managers, framed as efficiency tools, and implemented without much public debate. That is how change arrives unplanned.
What this means for health systems, professional bodies and educators
For workforce planners, the first step is visibility. Include allied health professions explicitly in AI workforce assessments, and track which tasks are being automated, where time is being saved and what it is being used for. The radiotherapy example shows how a single tool can equal one full-time post.
For department managers, decide what freed time is for before implementation, and put it in the business case. As we have argued about task-shifting, moving work between people or machines works when the system around it is designed for the change.
For professional bodies and regulators, update scopes of practice and competency frameworks to include AI supervision: validating outputs, editing automated contours, recognizing failure modes and escalating. Allied health professionals are becoming supervisors of machine outputs, and their training should say so.
For educators, build AI literacy and data skills into allied health programs now, alongside protected training in the unassisted skills needed when tools fail or perform poorly.
For vendors and buyers, evidence standards should apply equally. Tools used by allied health professionals affect patient safety as much as physician tools. Our piece on where AI diagnostic evidence is strongest applies here too: narrow, well-defined tasks are where AI performs best, and they are common in allied health.
Frequently asked questions
Will AI replace radiographers?
Unlikely in the foreseeable future. Radiographers position patients, operate equipment, manage radiation dose and care for patients during imaging and treatment. AI is changing parts of the job, including image quality checks and radiotherapy contouring, where NICE recommends AI-generated contours be reviewed by trained professionals. The role is shifting, not disappearing, amid existing workforce shortages.
What is AI autocontouring in radiotherapy?
Autocontouring uses AI to draw outlines of organs at risk on planning scans, a task that can take clinicians a long time manually. NICE’s 2023 early value assessment found it could be used while more evidence is gathered, provided every contour is reviewed and edited by a trained professional before treatment planning.
Is AI physiotherapy effective?
Early results are promising for selected patients with conditions such as back pain. Flok Health, a CQC-approved digital clinic, reported high patient satisfaction and symptom improvement in trials. Independent, comparative evidence is still limited. Patients with complex conditions, severe pain or neurological symptoms should consult a physiotherapist or other clinician.
How is AI used in clinical laboratories?
Laboratories use AI to pre-screen and prioritize digitized pathology slides, cytology and blood films, flag abnormal results and support quality control. In 2021 the FDA authorized Paige Prostate, the first AI tool for identifying suspicious areas on digitized prostate biopsies. Laboratory scientists increasingly review flagged cases and validate automated outputs.
Which healthcare jobs are most affected by AI?
Roles with a high share of structured, repeatable, image- or protocol-based tasks are most affected, including screening radiology, pathology, laboratory science, radiotherapy planning and some rehabilitation. Roles centered on hands-on care, complex judgment and relationships are less exposed, though almost every role will see some tasks automated.
The AI debate is fixated on whether machines will replace doctors. Meanwhile, the radiographer’s working day has already changed, and nobody has asked her what should happen next.
Which allied health role in your organization has changed most because of AI, and was that change ever discussed at board level?
