The AI in imaging paradox: expensive failure or necessary prerequisite?

AI for healthcare has generated enormous excitement and investment in recent years, and for good reason. It has demonstrated genuine measurable benefits in medical imaging, such as in workflow prioritisation, chest X-ray triage, cancer pathway acceleration and referral-to-treatment improvements. However, AI adoption remains at a relatively early stage of maturity for various reasons, and there remains a large gap between marketing messaging and day-to-day operational reality.

GE HealthCare’s Simon Rost recently participated in a plenary session on the subject at the UK Imaging and Oncology (UKIO) Congress 2026. This year’s congress covered how healthcare providers can place the human experience at the heart of diagnosis and treatment in an increasingly AI-focused age. The session saw a panel of experts discussing the nuanced successes and failures of AI in imaging, as well as the barriers that still need to be overcome before healthcare providers can use AI technologies to their full potential.

Meet the panel

Steve Holloway, Co-Founder & CEO, Signify Research

Simon Rost, Chief Marketing & Strategy Officer, GE HealthCare, Germany

Chad McClennan, CEO, Koios Medical, USA

Dr Rizwan Malik, Managing Director, SMR Health and Tech

Graham King, AI Special Focus Group Convenor, AXREM

Prof Catherine Jones, Consultant Radiologist, I-MED Radiology Network, Australia

The current status of AI imaging

AI in healthcare imaging has attracted significant investment and seen growing enthusiasm in recent years, yet widespread adoption remains a challenge. Many existing AI tools are too narrow and task-specific to deliver transformational change across healthcare systems, and the roll-out of newer, more universal technologies is notoriously slow. Healthcare AI requires extensive research, validation, regulatory approval and clinical evidence, but these long development cycles often conflict with investor expectations for faster returns and shorter adoption timelines. There are questions too about whether publicly funded AI programmes generate sufficient value relative to their cost and, as a result, some industry stakeholders are asking whether continued investment in new AI products is driving meaningful innovation or simply fuelling hype.

Part of the challenge lies in where investment and development efforts have been directed so far. Healthcare AI has largely prioritised clinically ‘exciting’, complex and visible applications, such as cancer detection and diagnostic decision making. These are undoubtedly valuable but often overshadow the development of solutions for more straightforward operational challenges. Lower risk functions that could deliver broader and more immediate benefits across imaging services have been under-prioritised, despite their potential for improving efficiency and reducing pressure in overstretched healthcare systems. For instance, AI for tasks such as scheduling optimisation, workflow management and referral guidance could maximise scanner uptime, reduce administrative burden, streamline patient pathways and free up valuable clinician time. However, these use cases often struggle to secure the same level of attention, organisational commitment and funding as those with a higher clinical profile. Healthcare organisations and industry partners should therefore identify genuine clinical and operational challenges first and then develop technologies that deliver the greatest value.

Addressing hurdles to adoption at scale

The barriers slowing uptake are evidently no longer purely technological; while many AI tools remain relatively narrow, immature and limited in scope, implementation challenges have also become a spanner in the works. Lengthy regulatory and governance processes – including repeated local approvals, clinical safety assessments and data protection reviews – are a major roadblock, as seen in the adoption of new AI tools for breast cancer screening. Alongside this, procurement cycles are increasingly complex, with compliance requirements and purchasing frameworks presenting substantial hurdles. Bureaucracy, limited accountability and an abundance of potential stakeholders only serve to heap on more layers of complexity, creating a state of decision paralysis. 

Another reason for the lack of AI maturity in healthcare is that successful pilot projects often fail to secure the further funding needed to convert them to fully-fledged, long-term, routine working practices. On paper, patient outcome is generally touted as the primary justification for AI adoption, with healthcare providers acknowledging that earlier diagnosis and faster treatment pathways matter more than productivity metrics. However, funding decisions are typically driven by measurable productivity and financial metrics because those are easier to quantify. Healthcare organisations often struggle to quantify the societal and clinical advantages of an AI tool and are dealing with inadequate data infrastructures. This means that quality improvements, reduced clinician stress and enhanced patient experience are typically overlooked as quantifiable, well-documented success metrics. Instead, boards and finance teams must make funding decisions based on available data, which has traditionally comprised improvements in productivity and economics rather than less tangible factors.

AI adoption in the NHS is a real-world manifestation of these complex hurdles and the uptake of new technologies for imaging applications has been mixed so far; significant progress has been made, but it is not yet realising its full potential. Internationally, the NHS is often regarded as relatively advanced in imaging AI, and its national scale creates opportunities for coordinated deployment that are not available in more fragmented and privatised health systems. Large AI pilot programmes have also helped to increase familiarity with the technology and build trust among clinicians. However, there are still persistent barriers and implementation remains slow, with repeated local assurance processes, complex procurement requirements and fragmented governance all contributing. The NHS has attempted to create some excitement and momentum around AI use in imaging, but it needs stronger coordination and streamlined processes to scale up existing infrastructures. It is clear then that although transformative AI technologies may exist, there is a pressing need to develop more standardised metrics for tracking patient-centred and health outcomes to support their widespread adoption.

The ideal model for clinical practice

There is also much more to learn about where in the healthcare pathway AI solutions are implemented, and how they are supported once they are. The preferred model for the future of AI is not for technology to replace clinicians, but rather for it to augment and streamline clinical decision making, for example, by identifying findings that clinicians could potentially miss. Consensus is that human expertise will remain essential, as clinicians can identify context that AI cannot understand, particularly in applications such as breast screening, chest X-ray interpretation and quality assurance.

The conversation around liability may consequently also begin to change as more AI is integrated into routine practice. Rather than focusing solely on the risks of AI producing an incorrect result, healthcare organisations may increasingly need to consider the implications of not using AI tools that have been shown to improve clinical performance and patient care. Choosing not to implement validated AI technologies could become more difficult to justify as the evidence base continues to grow. Patient expectations may accelerate this, as some already view AI as a valuable complement to clinical expertise, reinforcing the case for a physician-plus-AI model in imaging workflows: AI should be used alongside clinical expertise when supported by appropriate evidence and governance. It additionally calls into question whether the healthcare system is failing the patient by neglecting to use AI where available and delaying AI integration through lengthy procurement processes.

For these technologies to be implemented safely however, training and AI literacy are critical, but training in new technologies is one of the most overlooked areas of AI adoption. Organisations often deploy tools without sufficient preparation, and day-to-day operators frequently receive little practical education and ongoing instruction or guidance. Digital literacy is inconsistent and users are expected to teach themselves new skills on the job while juggling heavy workloads. Typically, no extra learning time is allotted, meaning that AI proficiency moves further down the list of priorities and placed on the back burner. Recommendations include investing in long-term, hands-on training programmes with continual instruction and user support. Training should explain why certain AI tools are being used, what they do and how their data should be interpreted in order to get the most out of novel technologies.

Leadership and industry cooperation

Crucially, successful AI adoption depends as much on leadership as it does on technology. Organisational leadership, clear ownership, executive sponsorship and accountability are widely considered to be some of the strongest predictors of success. Even the most effective AI solutions will struggle to gain traction without visible support from senior leaders and clinical champions. Successful deployments have commonly been characterised by strong leadership, clear decision making and active clinical engagement. On the other hand, unsuccessful initiatives often become mired in bureaucracy, with unclear responsibilities and a lack of organisational backing.

Clinical buy-in is also essential, and sceptics can become some of the strongest advocates once they experience the benefits of AI first hand. To encourage this, users need to be involved as early as possible in the implementation process, helping to shape deployment decisions and ensuring that solutions address real clinical needs. Greater national coordination is also important, particularly within the NHS, so that regional or local healthcare organisations don’t have to repeatedly solve the same governance, assurance and implementation challenges independently. Combining strong leadership, early user engagement and more coordinated approaches across health systems will better equip stakeholders to roll out AI at scale and realise its full potential.

At the same time, it is important for healthcare systems to strengthen their ties to industry, as the co-development of AI solutions with external stakeholders reduces risk. Industry increasingly sees itself as a partner rather than simply a supplier and, in many cases, commercial partnerships are driving innovation. These collaborations include advisory boards, co-development initiatives, research partnerships, acceleration programmes and managed service approaches. Healthcare-industry partnerships are becoming increasingly important, and the value of these relationships should not be underestimated. Concerted efforts must therefore be taken to foster ties with organisations outside of the healthcare sphere.

What needs to happen next

AI in healthcare imaging has come a long way, but much still needs to be done. Individual algorithms have been shown to be highly valuable for imaging in terms of improving workflow efficiency, supporting clinical decision making and contributing to earlier diagnosis. However, the conversation now needs to shift beyond technology itself to ensure that the benefits of AI translate into meaningful, scalable improvements to patient care. To achieve this, healthcare systems must place greater emphasis on measuring what truly matters. For example, rather than focusing solely on cost savings, productivity gains or return on investment, organisations need stronger evidence on clinical impact, quality of life and long-term patient outcomes. Better frameworks and data infrastructures are required to assess these factors and to more clearly recognise improvements in care quality, patient experience and clinician confidence.

The importance of building AI literacy across healthcare also cannot be underestimated, as clinicians, healthcare leaders and decision makers need the knowledge and confidence to evaluate, deploy and use AI effectively. Successful implementation therefore demands ongoing education, change management and support in the long term, rather than just at the initial launch. Perhaps most importantly, future AI adoption must start with the clinical problem rather than creating a technology simply because it is possible or available. This means that healthcare organisations should work backwards, identifying current patient needs and then asking how AI can address these specific real-world challenges, rather than implementing highly niche solutions or those that are not strictly necessary.

Ultimately, healthcare imaging has moved beyond asking whether AI works, and technological capability is no longer the biggest barrier to progress. Safe and large-scale adoption now depends on effective governance, leadership, evidence, training and implementation frameworks, as well as more and stronger ties between healthcare and industry. Established and dynamic partnerships are essential to developing, implementing and scaling AI solutions that address genuine clinical and operational needs. Overall, AI most certainly has the potential to improve imaging services and deliver better outcomes and experiences for patients across the healthcare system, but the next steps will be crucial in enabling the industry to take full advantage of these transformative technologies.