The role of a therapy radiographer in the age of Artificial Intelligence (AI)

Will the critical shortages of therapy radiographers mean that we are about to be replaced by AI, robots and machine learning systems and that will essentially solve the training, retention and employment problems in our profession by stealth?

The Society of Radiographers have just announced that the new apprentice programs are now “GO” and where a more vocational training environment in combination with a prospective employer and a degree course will allow employers to “attract and select individuals they believe have the potential to become radiographers”. It has also been announced this month that the University of Portsmouth is to close its degree course in radiotherapy and oncology in 2020 for which the timing is particularly ironic and may well impact on recruitment in the South further exacerbating the current problem.

I looked at these issues in my January blog and reported on some items in the media relating to this. The College of Radiographers published some of their latest feedback and information on Radiographer Apprenticeships in my February blog and now having read some of the latest books on the impact of Artificial Intelligence on us and especially the workplace, I thought it would be interesting this month to see how this might impact on our profession.

My question is that are the powers that be simply focusing on recruitment and retention issues and dealing with those when they perhaps should be concentrating on the role of a therapy radiographer and its definition over the coming years and how that might look in the years leading up to 2050.


Yuval Noah Harari’s predictions in his latest book… 21 lessons for the 21st century state that by 2050 not just the idea of a “job for life” but even the idea of a “profession for life” might seem antediluvian (belonging to the time before the biblical Flood) or just plain outdated.

He argues at the start that the masses now fear irrelevance in the face of the impending AI onslaught and that we are using our remaining political power before it’s too late in reference to the rise of Trump, Brexit and Nationalism. The technological revolution could push billions of humans out of the job market, leaving a jobless underclass and leading to political upheaval that no ideology will be able to handle.

He states that there is no clear picture of how the job market will look in 2050 but machine learning and AI will transform almost every area of work. He states that jobs that require specialisation in a narrow range of routine activities will be automated, GP’s could morph into an “family app” on a smart phone however the human care industry would thrive but he doubts that “Nurse robots” will ever arrive but elderly care will be a huge job-growth area.

The job market in 2050 will likely focus on Human/AI interaction roles for example a drone pilot and with humans servicing and leveraging AI to remain employed. Already data algorithms are the biggest buyers of bonds, shares and commodities on the stock market and in advertising the largest customer of all is the Google search algorithm via websites designed to pander to this and not actual human requirements.

He concludes that algorithms open us up to a rise in digital dictatorships but with economic safety nets in place, strong community initiatives and meaningful pursuits engaged then being made redundant by algorithms may be a blessing in disguise. I hope so!

“How will AI affect us as workers” asks Max Tegmark in his new book…Life 3.0 being Human in the age of artificial intelligence.

He believes we need to grow prosperity through automation without leaving people lacking income or purpose so we have the potential to create a fantastic future. He refers to a “Digital Athens” where robots replace the slaves of antiquity that left the Greeks of the past able to enjoy democracy, art and games without stress, the future vision would create a digital utopia and an abundance of consumer products that we have yet to fully envisage.

He goes on to look at what career advice we should give to our children now focusing especially on ones where machines are presently bad at performing the roles and may not get automated any time soon.

If you answer yes to these questions below more than once you are on the right track:

Do you have to interact with people using social intelligence?

Are you creative and come up with clever solutions?

Do you work in an unpredictable environment?

Teacher, Scientist, Entrepreneur, Lawyer, Social Worker, Engineer, Programmer and Hairdresser to name a few!

He suggests that if you work in a field of highly repetitive or structured actions then your time is likely up such as train drivers, tele-marketing persons, warehouse and bank workers to even car, bus and taxi operatives with accountants somewhere down the line (perhaps not far enough some might say).

The point is raised that perhaps one day all humans will be unemployable. Optimists say that automation creates many new alternative jobs whereas the pessimists suggest the law of supply and demand will reduce salaries to levels way below the cost of living.

Might AI lead the way in all intellectual tasks so that any remaining jobs will be low-tech? In the US in 1915 there were 26 million horses but in 1960 just 3 million and so that is a concern as the new jobs for horses never materialised. But in the new millennium due to unprecedented development of human/equine care systems including fun, sports and companionship the number has tripled to £9 million so there is hope.

He concludes that in a low-employment society a universal income support system added to increased activity within larger social networks, higher self-esteem, enhanced wellbeing and by being paid for simply making a difference will allow future societies to flourish through the greater profits generated from the AI revolution.

Automation in radiotherapy, a simplified timeline from my perspective and how this has impacted on our profession over time.

The early analogue Linacs had manual wedges and each beam was essentially applied by eye to hand co-ordination without computer control with radiographers running in and out of the bunker around 200 times per day. The treatment plan was delivered using pens, markers, rulers, calipers, tattoo ink and a print on piece of paper. The machines were not really iso-centric either. Port/Check films where simply Kodak plates for MV imaging that we rushed to and from the dark room post exposure. Patient management and administration systems were colour coded wall charts.

The advent of motorised asymmetric jaws and computer control systems allowed auto-sequencing of truly iso-centric beams so that shoe leather was saved and patients treated with less one to one interaction between patient and radiographic staff.

Patient management systems are developed that allow a treatment plan, now 3D in some aspects to be delivered automatically by the Linac’s computers along with storing the patient’s demographics and data online.

MLC development, enhanced control software and faster computers allow the development of IMRT and even less one to one interaction, you simply set the patient up, leave the room and return to remove them.

Unwanted patient motion now becomes an issue as we now don’t go into the bunker unless we want to take the patient off the couch and so the integration of cone beam CT and on-board imaging systems (OBI) allows for IGRT and a level of adaptive radiotherapy is possible.

Acronyms in radiotherapy become an industry leader and warn us of things to come.

We have even less time to stare at CCTV screens to check the patient is still where we left them as patients relatives keep disturbing us and there is far more administration to deal with on a daily basis. Treatments are far more complex and so motion tracking systems based on fiducial markers, infrared lighting/cameras or diodes and other technologies are developed that enables the computer to check the patient is where they should be and not the radiographer.

I recall asking a patient at Mount Vernon to keep still and said “off we go then” and for them to walk out of the room behind me when about to hit the “Go” button. CCTV and motion tracking don’t help here.

The crucial ability to check the patient remotely means that more effort is put into dynamic online imaging and so the idea of putting a Linac and CT scanner together is first mooted. A stand-alone CT and stand-alone Linac joined in the room by a couch on rails that automatically travels between the two systems works in a compromising way providing the patient doesn’t move!

A CT Scanner is then fully integrated with a Linac situated in the rotating component of the CT so that there is no compromise, it’s named Tomotherapy, image quality is good and adaptive radiotherapy possible but largely on an inter-fractional basis.

Someone then suggests that an MRI scanner and Linac should be integrated to allow intra-fractional and fully adaptive radiotherapy based on the ability to see the patient’s anatomy in real time. They are deemed to be mad and sectioned within the RT community (similar to my comments to radiotherapy managers in the 90’s that all planning would be carried out using CT Scanners and not simulators for which I was black-balled!).

Systems are then designed that allow the patient to be set up automatically as long as they are in an approximate position on the bed by taking a 3D image or shell of the patient’s external anatomy that is correlated to their internal anatomy by reference to prior CT data and the treatment plan. This also allows patients to breathe or not breathe depending on their diagnosis. These systems also negate the need for the rulers, pens, markers, calipers and even tattoos from initial paragraph. (Note: A tattoo is a pin-prick from a needle dipped in ink for those reading this who are not involved in RT).

Control systems now allow for machine gating so that treatment can be momentarily stopped while the patient or patient’s anatomy is put in the correct place based on a variety of technical factors such as how much water they have drunk prior to treatment.

Robots start to appear in main-stream radiotherapy with the advent of stereotactic radiosurgery and radiotherapy techniques that allow for 6 degrees of freedom. Their robotic design, coupled with real-time imaging, enables these systems to deliver a maximum dose of 6MV photon radiation directly to the tumour from many different angles with sub-millimetre precision. It does this by tracking and adjusting for tumour or patient movement during treatment to minimize radiation exposure to healthy organs and tissues.

Proton Therapy facilities with fixed-beam treatment rooms where the patient is rotated and translated in space with an industrial robotic arm that enable beam incidence from various angles have been developed with the patient treated in a seated position or even in bed. A combination of robots can hold the online imaging equipment, the chair or bed, the heavy collimators and also assist in positioning large devices, such as water phantoms for dosimetry and quality assurance measurements.

Robotic treatment couches start to replace standard ones that allow for accurate and remote geometric corrections of any misalignments. These are detected by state-of-the-art on-board image guidance systems that allow for precise corrections in six coordinates for IMRT, IGRT, VMAT and SRT tumour targeting.

The people who said that an MRI based Linac was possible are now not considered mad anymore and an MRI Linac becomes a clinical and financial reality allowing for truly adaptive radiotherapy on an intra-fractional basis with not one but two manufacturers (Elekta and ViewRay) installing systems world-wide. Therapy radiographers need to learn fast from their diagnostic peers that the magnet is always ON!

Pic: Elekta Unity MRI Linac

The ongoing development of cone beam CT also now allows us to mount a Linac on a rotating ring with integrated cone beam CT and OBI so that a Linear Accelerator looks increasingly like a CT scanner but isn’t and can deliver fast, image guided treatments and plenty of Acronyms!

Lastly, while under IRMER regulations we need to ensure we are treating the correct patient and the correct part of the patient there are paradoxically new systems that can identify the patient automatically for us using a facial scan or palm reader. These ensure we are treating Mrs Smith and not Mr Jones and that the plan delivered is in fact their own but the paradox is that a very large number of Linac bunkers still rely on CCTV systems to check for unwanted patient movement and radiographers staring at the TV screen all day which is not that efficient or accurate anymore.

New patient ID systems can integrate patient identification, accessory verification, surface guided positioning & monitoring and breath-hold with the patient management systems and oncology information systems. This means that treatment and administration are now largely one simplified and completely automated process far removed from the colour coded wallchart at the start.

Treatment planning automation

Meanwhile in the treatment planning world I recall hand planning using isodose curves printed onto tracing paper and combing these beams to create a 2D plan, even new Linacs came with large binders full of isodose curves for every possible field size and wedge orientation. Learning the basics was the key to implementing automated systems.

Treatment planning quickly goes from 2D to 3D using faster computers and more accurate algorithms taking into account inhomogeneity in three dimensions and more accurately calculating the effect of photon interactions often at quantum levels while these complex plans are able to be delivered directly on the patient with seamless integration into the Linac control systems.

Auto-segmentation and rendering of critical organs and tumour volumes becomes standard practice as well as full inverse planning whereby we dictate the parameters required and the planning system and Linac deliver a fully optimised course of treatment.

The contemporary planning systems allow modern Linacs to deliver complex treatments such as IMRT, VMAT and SRT automatically in real time with little radiographer interaction.

The reality now is that fully adaptive radiotherapy allows the treatment plan to be modified on the fly during beam on based on the real time position of critical anatomy and the direct input of a clinical oncologist further removing the radiographer from the coal face.

Proton based CT Scanners will likely be the holy grail for truly adaptive Proton Beam Therapy treatment and planning in the future and that incorporate Proton interactions at quantum levels too.

Machine Learning according to Wikipedia

Machine learning (ML) is the scientific study of algorithms and statistical models that computer systems use to effectively perform a specific task without using explicit instructions, relying on patterns and inference instead. It is seen as a subset of artificial intelligence. Machine learning algorithms build a mathematical model based on sample data, known as “training data”, in order to make predictions or decisions without being explicitly programmed to perform the task.

Machine learning algorithms are used in a wide variety of applications, such as email filtering, and computer vision, where it is infeasible to develop an algorithm of specific instructions for performing the task. Machine learning is closely related to computational statistics, which focuses on making predictions using computers. The study of mathematical optimization delivers methods, theory and application domains to the field of machine learning. Data mining is a field of study within machine learning, and focuses on exploratory data analysis through unsupervised learning. In its application across business problems, machine learning is also referred to as predictive analytics.

Machine learning in radiotherapy

RaySearch, one of our RadPro partner companies now offer a smarter, more efficient clinical reality with the world’s first treatment planning system with machine learning.

High-quality treatment plans are created automatically using machine learning models trained on hospital data or data from outside leading cancer centres.

Organ segmentation is carried out in less than a minute it seems while machine learning transforms every new byte of data into insight to improve future treatment for patients. They claim that it also brings a new dimension to automation, for planning that’s faster and smoother than you ever imagined.

Pic: RayStation image


Key features are:

Generate contours of organs in less than 45 seconds with deep neural network models

Generate personalized treatment plans in minutes

Benefit from trained models from leading cancer clinics

Train your own models

Share models with other clinics

And finally, the Orwellian outlook for radiotherapy

Out-of-Treatment Room Set-Up for Patient Positioning in External Beam Radiotherapy

In 2009 in Sweden the use of a transport system for out-of treatment room set-up in external beam radiotherapy was initially explored (L. Weber, K. Westerlund, I. Näslund, and H. Dahlin et al)

They looked into the vibrations which a patient is subject to during movement of the trolley as well as the accuracy achieved when using a reclining table for patient set-up and immobilization.

The standing patient leans back and the table slowly reclines leaving the patient supine and they argued that use of a reclining technique improves patient positioning. A vacuum immobilisation bag was also used and located on the trolley prior to patient loading.

They concluded that the reclining technique also allowed a more flexible approach in the department design as the patient loading station could be located remotely and the patient transported between the radiotherapy station and the dressing area, or optionally pre-verification, thereby allowing higher patient throughput or alternative target verification methods.

The trolley is driven by electrical motors, one on each wheel, powered by a rechargeable battery and its top is essentially a heavy-duty, rolling sheet.

In recent years it has been further developed so that in order to improve patient throughput in a busy RT centre or PBT facility that a series of tables are stored in a “patient loading area” and then the table follows an electro-magnetic tracking system located in the floor and can transport the patient between Linacs or PBT gantries, planning scanners and other key parts of the centre automatically and this became a commercially available product.

How such a system if fully developed and implemented would impact on us as therapy radiographers is a moot point where-by we might simply only need someone to “load” the patient before AI takes over the entire patient pathway each day?

 Pic: George Orwell taken before 1984


The potential future impact of AI on radiotherapy– some interesting quotes from other sources

“Overall, an ideal (AI) system could substantially speed the process, reduce the time burden of human intervention”

“In the future, it is likely that advanced computational techniques will enable such rapid and reliable automation as to reshape resource utilization and staffing levels, training requirements, reimbursements, and patterns of care, as well as other aspects of Radiation Oncology”

“It is also likely that the most successful implementations will require human engagement to utilize AI to maximally augment human skills. Indeed, technological advances are often accompanied by loss of some clinical or technical skills with gain of others”

“Future trainees and current practitioners alike will need to understand the applications and limitations of multiple AI tools of the future. Current critical skills such as contouring will see their importance fade over time, to be replaced by a fluency with AI and novel human–machine interfaces”

“AI is sure to affect treatment capacity, treatment capabilities, and safety and QA frameworks, with significant repercussions for patients, providers, and the healthcare system as a whole. The international Radiation Oncology community will need to work together to ensure optimal utilization and coordination of talent, training, investment, and resources in order to maximize the potential of AI”

“Even as it may seem part of a more distant horizon, the opportunity cost of ignoring AI at this juncture is steep”

My conclusion

Will we as therapy radiographers create an employment niche for ourselves in the bright new AI world by 2050 or will we by stealth be replaced by robots who simply put patients into an automated radiotherapy treatment production line and a big-data algorithm, AI-based sausage factory?

Hopefully this blog is thought-provoking and so please send me your comments to or our twitter and facebook feeds.


Next month – June. Creating the first truly mobile radiotherapy service in the UK