Artificial intelligence has been granted a newfound focus in recent years, as rapid advancements in generative AI raise questions about how it can be used across professional fields. But there is more to AI than responding to questions or creating images – its potential to improve the efficiency, effectiveness and quality of radiography cannot be underestimated.

The NHS England Framework for Allied Health Professionals recognises that digital health solutions can improve quality, safety and efficiency.

It explains: “Compared to other AHPs, radiographers will increasingly be exposed to some of the more advanced developments in digital diagnostic capabilities through the evolution of artificial intelligence and machine learning innovation in clinical decision support.”

In 2022, meanwhile, Unity Insights and Health Education England published a study which found 240 different AI technologies had been deployed in the NHS. Of these 240 technologies, 34 per cent were being deployed in diagnostics, and of that 34 per cent, just under half (49 per cent) were involved with imaging solutions.

This roadmap identified the top workforce groups with the most potential to use AI (by percentage of applicable technologies) as:

Human oversight

The Society of Radiographers has released official guidance for the past few years surrounding the importance of implementing AI technology in a safe, efficient and effective way.

It warns: “AI is being rapidly integrated into imaging equipment with arguably little consideration as to how it influences radiography practice and frontline services. Current regulatory frameworks stipulate that all AI systems that are deployed clinically require human oversight of their implementation. 

“This is potentially adding to the radiographer’s role and increasing the need for a high level of digital fluency among radiography staff as they must learn to evaluate, interact and oversee the actions of AI driven tools within their workflow.”

The Society calls for research to investigate the impact of AI on: quality of service, patient care and radiographers’ roles and working practices. 

It adds there is a need for ‘real world’ clinical validation of AI interventions.

Tracy reiterated that while development of many different kinds of AI technologies remained ongoing, the principles of the SoR policy statement on AI are an important strategy. The SoR aims to influence and direct improvements that can be made by implementing new technology in healthcare, locally, nationally, and internationally, and to promote safe and ethical practice that will enable better patient service user outcomes.

She said: “It’s got to be safe and effective, and also consider if it is economically effective? What is it we need to assist clinicians and patients?”

The Society’s AI Advisory Group has also released a series of webinars, policy and guidance documents for radiographers interested in getting to grips with the technology, its use, and its potential impacts on the profession.

Alongside policy on ensuring members are consulted at every stage, SoR guidance states:

• Radiographers should will take into account the opportunities for innovation and development of practice and patient care that are afforded by evolving technology, AI, machine learning and deep learning
• The guidance development was mindful of opportunities to adopt AI systems for the improvement of its service to the membership, and new initiatives will be tested and assessed to strike the right balance between additional efficiency with appropriate transparency and member engagement
• The protection of patient data is paramount

The guidance adds: “We would expect that as a minimum standard, any AI system will have clinical governance measures in place and will be compliant with current Quality Standards for Imaging and UK GDPR.”

Quality assurance

Avneet Gill is a diagnostic radiographer currently undertaking a PhD at Ulster University focusing on potential new applications of AI. She sat down with Synergy to discuss her research.

She explained Quality Assurance is one area AI can help radiographers with – many NHS Trusts now use Computer-Aided Diagnosis (CAD) alongside their imaging, which can “run in the background and wean out the easy ones,” Avneet said. This software processes images to identify typical appearances and highlight conspicuous sections.

CAD AI usage varies across the country, and is currently used mostly for pathology detection.

AI can also be used to advise on image optimisation, Avneet continued, by informing the radiographer if the patient is positioned correctly or if a scan should be repeated. A similar technology can track patient positioning and cease scanning if a patient moves out of position.

Avneet added: “Our workforce is trying loads of different things to support the lack of radiographers and apprentices. There is a shortage, but AI is a way to support that – it can improve the detection of pathologies, it can improve the amount of scans being reported. It is a positive thing.”

Avneet described a novel AI technology that sought to aid in triage, sorting scans by severity for radiographers to report.

“There's so much potential for where it can go. The breadth and scope of a radiographer’s responsibilities is going to be expanded to include most modalities. The new registrants are going to need to understand everything. How are we going to do all that without help?”

Claire Currie, senior lecturer in diagnostic imaging and part time PhD student at Glasgow Caledonian University, agreed the scope of radiography is growing, and there is a need for support to meet this extra capacity.

Demand for radiology services is only growing, driven by an ageing population, changing clinical pathways, new guidelines, and increased screening to support early diagnosis.

“It is now urgent that strategies are put in place to ease this disconnect,” she added.

One way this might be achieved is through a clinical decision support (CDS) tool, which aims to eliminate inappropriate referrals and missing request information. CDS systems guide clinical decision-making, and was one of the earliest forms of AI in radiology departments.

Claire believes there is potential for CDS to be updated and advanced, for more personalised support.

CDS systems work by connecting signs and symptoms to diagnoses and imaging tests, as well as rules and guidelines with actual patient data, such as case symptoms and clinical history, before producing a radiology referral, and presenting the clinician with the most up-to-date rules or guidance.

“A CDS can improve the quality of patient care foremost, followed by potential workflow efficiencies as well as efficient management of resources,” Claire said.

“For the quality of care, CDS could ensure that patients receive the right test the first time, allowing or enabling clinicians to provide or take patients to the best treatment pathway and, by extension, minimising any necessary radiation to the patient.”

Safe and effective?

Tracy explained that questions of evaluation are important to answer now, to ensure effective implementation and governance of AI into the future.

“What will the research or service evaluation approach be? What metrics will be measured and will findings be transferrable to other sites and for other use cases? Is it safe and effective? Is it going to make a difference to the patient pathway, to patient outcomes? How is it going to affect the staff and the roles that we undertake?

Tracy described some AI use cases tackling problems that are unlikely to make a difference – AI that checks whether a knee image is reversed, for example, should have little practical impact, because radiographers are trained to check key reversal markers during image evaluation.

Avneet, meanwhile, warned people are resistant to the use of AI because of the risks of it becoming a “black box”, with the users unable to understand how it works.

Evidence also suggests people are more likely to favour advice from automated decision-making systems, even when contradictory information is provided from a non-automated source.

“People do rely on it maybe too heavily, and then they get into trouble and it leads to more of a negative perception of AI,” she said.

“There’s also some confusion from radiologists and reporting radiographers around who takes legal responsibility. If an AI algorithm says, ‘this chest X-ray has lung cancer’, and the department follows treatment for that, but it turns out that wasn't correct, then someone has to be to blame. 

“That patient might have undergone extra treatment, or extra imaging, or extra radiation they didn't need. People are a bit scared of that legal aspect of it. The amount of products out there which haven't been tested on a really large dataset is quite astounding.”