Therapeutic radiographers now deliver MR-Linac treatments without clinical oncologist present

Is this an opportunity or a threat to our role and will advances in AI allow us to reach the holy grail of therapeutic radiographer led Magnetic Resonance guided Radiotherapy?

I must admit that when I saw this headline above on the internet and another one announcing ‘Radiographers deliver on-line adaptive radiotherapy without oncologists’ in Synergy, the magazine provided by the Society of Radiographers, I immediately sat up and listened! I wanted right away to find out more about this technology, write a kind of ‘Magnetic Resonance guided Radiotherapy or MRgRT for Dummies’ blog and see exactly what was behind this headline and why it stood out so much to me?

The thrust of the articles I read suggested that:

‘Therapeutic radiographers in The Royal Marsden NHS Foundation Trust’s MR-Linac unit have started treating patients for the first time without a clinical oncologist present, following a training and supervision programme’

The MR-Linac, as I will discuss below allows treatment to be adapted to the position or shape of the target volume and organs at risk (OAR) in real time and is now generally known as MRgRT. Over the past three years and since the first patient was treated on their Unity System this has required a clinical oncologist to be present in the control room.

Therapeutic radiographers while always historically responsible for patient positioning, imaging and treatment delivery have now taken on the additional role of ‘contouring’ at the RMH in part due to comprehensive additional training and also with the tacit approval of the oncologists for treating cancer of the prostate and also some well-defined metastatic lesions to date. Physicists are also present for online treatment planning and dosimetrist validation while extending the radiographer’s role to cope with more complex tumours sites is an ongoing learning and training process it seems.

Pic: The Elekta Unity MR Linac just like the one located at the Royal Marsden Hospital

The role of a therapeutic radiographer from the day I started as a qualified member of staff until now (aside from some unusual outliers that I will come to later) has always been that we deliver radiotherapy without an oncologist being present. That is our primary role and ‘raison d’etre’. We usually have to refer to an oncologist if there is a problem, or if the goalposts that we work to with regards patient or tumour position or motion have moved outside pre-agreed limits when imaging the patient but not in real time and online, until now.


However, new technology like the MR-Linac have not only moved the goalposts but taken them away! The ability to adapt the treatment dynamically in an on-line mode and in real time is a new and disruptive technology (disruptive in a good way, see below) that we as therapeutic radiographers will be required to adapt to ourselves through intensive and dedicated training but also, I believe by embracing AI.

So what is a ‘Disruptive Technology’?

‘Disruptive technology is an innovation that significantly alters the way that consumers, industries, or businesses operate. A disruptive technology sweeps away the systems or habits it replaces because it has attributes that are recognizably superior’ – courtesy of Investopedia.

My first adaptive treatment!

When asked by one of my Middlesex Hospital ‘Consultant Radiotherapists’ to treat a patient’s whole brain palliatively giving 20Gy over a course of 5 days using a 250kv applicator meant ‘knowing where to put it’ using anatomical markers. I had only just qualified and this was my first treatment machine posting. She handed me a signed sheet of paper or treatment sheet, the prescriptive dose, an X ray and off I went. She didn’t sit there while I aligned the machine, chose a suitable applicator and turned it on each day, largely because I was fully trained and accredited to do this. The ‘adaptive’ part was some sticky tape applied daily to keep the head still and stuck to the applicator for each side so it didn’t move. Those were the days however I’ve been criticised for being a little stuck in the ‘80’s and so let’s move on quickly.


The headline issues above really come down to ‘contouring’, a routine that has probably changed little in the past 40 years until now. Initially a treatment planning system or TPS would provide a light pen or tracker-ball and then later a mouse so that you could manually contour anatomy on a CT scan and submit that to a dose calculation algorithm within the TPS.

Fast organ rendering based on voxel differentiation/contiguous voxels or simply based on Hounsfield numbers automated what I always referred to as a ‘bulk’ outlining mode to save time but when the target volume required delineating, this was always the clinical oncologists place, as now.

The Precision Therapy Render Plan 3D system was the first ever commercially available planning system that offered fully automated organ or volume rendering in the mid 90’s. By simply locating the mouse inside the region of interest and clicking, the system would automatically run through the CT slices and come up with a 3D model of the lungs, bones, bladder, oesophagus etc in real time but other more complex anatomical structures would have to be contoured manually in those early days. You needed to have a good contrast between the CT numbers of the areas to be rendered and those nearby for it to work reliably. At least calculating a DVH for the lungs for instance was now far more simplistic and far less time consuming.

Now novel state of the art AI algorithms and deep learning technologies can largely automate the contouring process within treatment planning systems and now the MR-Linac and use software that can recognise and reproduce them on a daily basis using advanced segmentation techniques.

Deep-learning based auto-contouring…. what is deep learning, here are two popular definitions?

Deep learning is an artificial intelligence (AI) function that imitates the workings of the human brain in processing data and creating patterns for use in decision making. … Also known as deep neural learning or deep neural network – Investopedia

Deep learning is a class of machine learning algorithms that uses multiple layers to progressively extract higher-level features from the raw input. For example, in image processing, lower layers may identify edges, while higher layers may identify the concepts relevant to a human such as digits or letters or faces – Wikipedia

Pic. Machine learning in action. If you present an image of a dog, ML algorithms access vast amounts of learnt data to enable them to distinguish the actual breed, in this case a ubiquitous French Bulldog!

Manually defining OAR’s is time-consuming and its major drawback is that it suffers from inter and intra practitioner variability and so deep learning models can automate this process, saving us lots of time and effort and provide enhanced accuracy and verification.

Deep-learning segmentation models ‘learn’ from vast numbers of prior patients and their data anonymously and automatically, generating contours of all relevant regions of interests and OAR’s in just a few minutes in Cone-Beam CT, conventional CT and also MR image data sets. However, deep-learning segmentation is not designed or intended to identify tumours presently but tumour segmentation using AI is now under active research and development and so will take treatment and planning to yet another level at some stage in the near future.

Deep learning works within what is known as a ‘neural network’, where each image voxel belongs to unspecified tissue or to a specific structure. It can be set up to work on different body sites and imaging types. A neural network is commonly defined as a ‘series of algorithms that endeavour to recognize underlying relationships in a set of data through a process that mimics the way the human brain operates. In this sense, neural networks refer to systems of neurons, either organic or artificial in nature’.

Deep-learning segmentation models within planning systems can also be ‘pre-trained’ based on any international treatment centre’s individual protocols and therefore simplifies in house training, accuracy and performance of the contouring process at each site.

It also allows for more closely defined agreement between peer to peer members of the multidisciplinary team or MDT either working in the treatment planning process or delivering adaptive treatments as looked at in this blog.

Perhaps eventually AI will act as the ‘arbiter’ for what are often subjective decision-making variations in the contouring process. After all there are now commercially available QA systems that are designed just to specifically check the accuracy of segmentation techniques, its algorithms and software so most areas of potential error are now well and truly covered!

In the case of the Elekta Unity, Elekta’s has said that is has used ‘thousands of MR image data sets’ provided by their MR Linac consortium and have used these to ‘train’ their own auto-contouring algorithms that will doubtless help facilitate the clinical implementation of online and real time adaptive radiotherapy where the patient’s plan or treatment is adapted daily using, in MR-Linac parlance ‘ATP and ATS protocols’ that I will come to soon.

Machine learning and ‘intelligent’ treatment planning and delivery.

In the near future AI will very likely manage how we oversee complex radiotherapy processes with automatic contouring, essential for adaptive radiotherapy being the ‘first taxi of the rank’. Just as in Radiology, where machine learning already automates diagnosing disease, these advanced algorithms will save time by automating plan generation and organ segmentation, enhancing accuracy in radiotherapy but also importantly allow therapeutic radiographers to return to the driving seat when delivering MRgRT and so why was ‘contouring’ the main sticking point in the headline of this blog?

I thought therefore that I needed to learn a little more about how MR-Linacs work and the treatment process or workflow involved to implement them clinically. Please bear with me as I am no MR expert and so like many of us am involved in a steep learning curve but this is what I thought was the most relevant if slightly basic information.

‘MRgRT for Dummies’- Why is the Unity MR-Linac so different to what we are used to in radiotherapy?

To start with it’s based on diagnostic high-quality MR Images and the couch can only move ‘superiorly and inferiorly’ once the patient is being treated.

No breath-hold systems or techniques are possible at the time of me reading up.

There are two treatment modes, ATP and ATS.

Adapt to position or ATP involves a virtual couch shift or what is essentially an isocentre move to realign the patient’s position and tumour at the time of treatment with a quick online recalculation and verification. While re-calculation presently is generally undertaken by a dosimetrist, this could readily be made a role for a therapeutic radiographer with limited additional training. As requisite treatment planning skills include treatment plan evaluation and assessment of critical errors, it is not a huge leap of faith to see us carrying out online ATP plan preparation and or recalculation.

Pic: My MR Linac blog does feel a little like something from the ‘Books for Dummies’ series, sorry!

Adapt to shape or ATS is far more complex and means that the target volume has changed shape or form and so the contours need review, recontouring or adaption and a complete re-plan is required while the proximity of OAR’s may also have altered when the ‘anatomy of the day’ is imaged and these need to be verified, adjusted and or recontoured too.

This is very time consuming under current protocols it seems and involves direct input of an onsite MDT of clinical oncologist, physics/dosimetrists and therapeutic radiographers to varying levels depending on what centre you work at or in what country you are based.

The dosimetrists involved in the MR-Linac process have different responsibilities in the ATP and ATS workflows and will also require enhanced training and skill sets for ATS operation while as above, do not need to be involved directly in the ATP process.

This is the workflow for a typical ATP Prostate treatment as far as I can see:

• Patient set up

• Pre-treatment MR imaging

• CT/MRI registration and approvals

• ATP planning and isocentre shift

• Plan/dose approvals

• Verification MR taken and re-assess plan

• Plan checking

• Proceed or re-plan and approve

• MR Linac treatment delivered

Real-time planning and complex adaptative processes will require therapeutic radiographers to embrace further training and by default extend their abilities to contribute to the ATS workflow in the same way as ATP. Once again, the amount of training will depend largely on country and centre where skill and knowledge levels can vary considerably.

There is an excellent recently published paper by Helen McNair on this subject that looks at the staffing, workflow and training issues with regards to MRgRT implementation at the Royal Marsden and other sites that have a MR Linac and is well worth a read.

Without a doubt the main thrust of this article is that via an enhanced national framework for specialist training, therapeutic radiographers will eventually deliver this treatment autonomously with other MDT colleagues available largely on an ‘on-call’ basis.

It is proposed that all therapeutic radiographers will need to have the skills to deliver this treatment and not just a select few and that learning ‘on the job’ is likely to be insufficient while student training programmes will need to incorporate the MRLinac workflow asap.

She concludes: To prepare for the future, training and education is key, and a national framework is required to prevent variable practice. MRIgRT implementation will increase and radiographers must be prepared to be actively involved in this radiography step-change.

Other MR Linac ‘workflow ways forward’ that came to my mind.

Another training scenario that might work would be that MR-Linac operation is an additional formal qualification to the current degree or future apprenticeship that might take 3 to 6 months to complete perhaps. The downside of this would mean some staff would not be able to deliver MRgRT unless formally qualified? Having an additional qualification on top of the DCR or degree is not unusual in our sector, I know many ‘dual-qualified’ or ‘higher’ peers but this concept for the MR Linac may be anathema these days. What do you think?

On call workloads would be reduced if ‘pop-up’ remote personal workstations are made available to ‘virtual’ clinicians and physicists so that they can have direct input in decision making, additional planning and verification but work remotely and or on the internet. However, I have read that operating on an ‘on-call’ basis is not ideal if the clinicians called upon are not proficient in a specific tumour site and MR-Linac workflow or in the case of planning, if a different physicist is sent for than the one attending on a previous treatment day.

If physicists and clinicians are able to be on-call then this would dramatically reduce the numbers of staff involved however in MRgRT there remains a lack of agreement worldwide regarding staff combinations required for ATP or ATS treatment delivery.

Either way, the move to radiographer led MR Linac treatment is firmly on the horizon.

Training and role development for therapeutic radiographers.

For therapeutic radiographers to perform clinical treatment decisions based on high quality MR imaging represents a very interesting and enhanced role development for sure while the clinical oncologist still has overall responsibility for the treatment even if they are on call or located in a virtual environment.

However, we should be wary as scenarios exist in other radiotherapy modalities that I will come to later that might be seen as ‘outliers’ where clinical oncologists and sometimes radiographers might not be present at all when treatment is delivered!

In the past, radiotherapy has had to put up with using what might be termed ‘inferior image quality’ when comparing portal imaging, CBCT, old CT-Sim systems and Sim-CT flat panel based products to our diagnostic colleagues but that is no longer the case as mixed modality, high quality imaging is seen as the future for radiotherapy.

It is fairly obvious to me now that for MR-Linac treatment to be the next radiotherapy panacea it needs to be led by therapeutic radiographers just like IGRT and CBCT based delivery is now but with additional requirements that are likely to include the contouring of targets and OAR’s and more complex planning and dosimetry knowledge and experience. But as above, AI is about to solve the contouring issues very shortly by default.

Pic: The Viewray MRIdian MR-Linac is the ‘other’ system on the market.

It is likely that enhanced training will encompass MR physics and safety, MR imaging and interpretation of anatomy CT/MR registration, MR-Linac based treatment planning, evaluation, checking and QA while a refresher in patient set up and treatment delivery for this specific machine would be also required but what does the Society and College of Radiographers say?

Will our diagnostic colleagues join forces with us to deliver this treatment?

The Society and College of Radiographers (SCoR) have acknowledged that the ‘needs of both UK therapeutic and diagnostic radiographers in terms of education and training in the use of MR in the radiotherapy pathway must be addressed’ as the roles are likely to become more intertwined in the coming years.

You could argue that by having diagnostic radiographic staff included in the MR Linac workflow (an obvious conclusion based on their knowledge of MR imaging) offers a threat to our role as a therapeutic radiographer as I said at the start but I also read with interest that complex contouring requirements for some hepatobiliary cancers for example, might have to include a radiologist’s direct input, further complicating matters and the clinical implementation/treatment of other cancers.

The SCoR released a document a few years ago that largely relates to MR safety issues and the potential training required but this will be reviewed in September 2021 and so it is likely to be a vastly different document now.

This guidance is divided into two main sections:

Recommendations to radiography educators for pre-registration education and training, either through the traditional higher education route or via an apprenticeship route.

Recommendations to employers and radiography practitioners to support the education and training of the existing workforce, both those working within the MR-RT (MRgRT) services and the wider workforce.

Aside from MR imaging safety and training they add this training diktat:

  • Additional learning and development for on-treatment MR-guided radiotherapy and understanding of potential side effects specific to the MR-RT environment
  • Ability to quickly and accurately contour/modify (at minimum) normal tissue contours for the purposes of adaptive radiotherapy; application of knowledge of the impact of dose on organs at risk;
  • Understanding of adaptive radiotherapy strategies, and of how and when to apply them.

The SCoR conclusion is shown below, you may have alternative views, if so it would be good to hear them:

There is a role for the entire radiography workforce in radiotherapy as MR in RT services are integrated within radiotherapy nationally. Therapeutic radiographers directly involved in the use of MR in RT, and all members of the wider radiography workforce, should have the knowledge identified above, with a clear understanding of the eligibility criteria for treatment and referral mechanisms in place across the UK.

Therapeutic radiography staff using MR in RT will need to develop further competencies to support patients through the pathway, to deliver treatment and integrate MR effectively into the patient pathway in order to support the goal of improved outcomes for patients, and to also ensure that therapeutic radiography workforce skills are effectively developed to enable full contribution

The outliers…radiotherapy carried out under IRMER guidelines but not involving radiographers or clinical oncologists directly!

Intra-Operative Radiotherapy or IORT for early stage breast cancer – This relies on an MDT located in theatre at the time of surgery. The surgeon removes the tumour under wide local excision protocols and inserts a spherical or balloon applicator where the tumour was once located. As long as a clinical oncologist’s signed treatment prescription is located in theatre then there is actually no reason why the oncologist has to be present just as they do not attend all brachytherapy sessions such as with HDR after-loading techniques.

Pic: IORT for early stage breast cancer schematic diagram.

The dose and treatment time can be calculated by a medical physics expert or MPE in theatre and they are equally able to turn the machine on and off and deliver the treatment whereby in some systems a 50kv miniature X-Ray tube is inserted automatically into the applicator. This is probably not widely known and while most centres who deliver IORT in the UK will always have an oncologist and radiographer present, it is not mandatory at all and may be one reason why IORT is often seen as ‘voodoo’ in the UK!

Gamma Knife – These treatments can be calculated and delivered by an MPE under IRMER without a radiographer being present.

Seed Implants of the Prostate and the TUI procedure – Another ‘outlier’ where a urologist positions needles and radioactive seeds transperineally into the prostate gland under ultrasound guidance and within a special template either as per a pre-planned layout or in an adapted way during the procedure whereby PSA levels in specific parts of the prostate can be used to adapt the needle and seed positions dynamically. An MPE is also located in theatre to assist and advice with the application of a preplan or calculate dose and evaluate needle and seed positions to create a post implant plan summary. The VariSeed system incorporates MRI fusion and PSA evaluations of biopsy tissue samples taken from planned seed and needle positions that allow a truly conformal plan to be delivered based on active tumour positions within the gland.

Other areas where therapeutic radiographers may be marginalised especially in the US are dermatology centres delivering radiotherapy and I know of some very small US sites whereby physicists and oncologists work together to treat patients at single Linac locations and so we should always be wary as a profession.

My conclusion

By embracing the rapid progress of AI that also automates the radiotherapy contouring process making it fast, accurate and verifiable, added to dedicated training programs and access to a novel virtual MDT environment, therapeutic radiographer led MRgRT is clearly within reach!