In this study, we found that children and young people were generally supportive of integrating AI into their healthcare journey, provided certain conditions were met. They expressed a strong interest in being informed about the accuracy of the AI tools being used, and conveyed a desire to be asked for consent for the application of AI on their medical images. Most children and young people expressed a preference for human oversight of the AI tool, though their willingness to accept AI assistance did not seem to vary greatly between different diseases as long as the outcomes were accurate. The respondents in this survey considered the accuracy of any imaging results to be more critical than the speed at which they were provided, in other words, respondents did not want accuracy to be compromised at the cost of speed.

Interestingly, respondents who had fractured a bone that was originally missed, were more likely to agree that AI could be more accurate than healthcare professionals at looking for bone problems on radiographs; whereas those who had fractured a bone (detected accurately) were the least likely to think that AI would be more accurate, presumably because they did not experience any particular delay or issue with their own diagnosis. This important finding highlights the need to ensure an appropriate mix of patient stakeholder engagement in any AI discussion, as personal medical and direct life experiences will affect perceptions about the potential for using an AI tool. We do acknowledge that the proportion of respondents in this study stating they had a ‘missed fracture’ may be higher than the general population (23.7% in this survey, versus an estimated 5–19% missed paediatric fractures by emergency clinicians (of which 5–28% are estimated to be clinically relevant) [30,31,32,33,34]. Emergency clinicians are usually the first healthcare professional to assess and discharge the patient, many times prior to formal radiology input in the UK). This higher rate in our survey could be for several reasons—those with missed fractures may have been more motivated to take part and many respondents found the link to our online survey via the ‘Brittle Bone Charity’ website, which supports patients with osteogenesis imperfecta. These patients have multiple fractures, many of which are missed and therefore their likelihood of a missed fracture may be higher than the average population. We know many children with this condition responded to our survey based on inference from free text comments, however due to preservation of respondent anonymity in our survey, we cannot quantify how many have this diagnosis.

There have only been a few publications exploring the opinions of children and young people on the use of AI for healthcare and none at present relating specifically to medical imaging. One study by Visram S et al [21] presented 21 members of a the wider GOSH YPAG with a variety of applications of AI in healthcare to understand areas they considered important for future adoption. Key themes surrounding governance, trust and human-centeredness were deemed important alongside patient empathy and safety. Another study by Thai K et al [22] interviewed 28 paediatric patients at a large urban children’s hospital and explored their opinions regarding the use of AI in clinical trials, clinical practice and health data research. A strong theme that emerged in this study was the need to maintain human interaction between patients and their physicians, although there were positive views relating to the use of AI for research and clinical care.

Whilst not AI-specific, there have been other surveys conducted on children’s views on the use of technology in general within healthcare [35], specifically for the use of robotics and virtual reality in hospital and educational settings [36,37,38]. In one scoping review looking at 73 publications relating to the use of robots in healthcare [37], it was found that the use of this technology was highly acceptable to children, parents and medical staff and feedback from robot usage was mostly positive. Although this type of technology differs from AI, it does support the acceptance that the younger generation have for interaction and integration of novel technologies for their own healthcare. In another review looking at 38 articles evaluating children’s concerns and needs in health technology (e.g. telehealth, medical devices, augmented reality) [39], four general overarching themes were found—issues relating to the stigma of using technology, data privacy, the trustworthiness of the technology and whether this was developed with age appropriateness in mind. Whilst the former may not be directly relevant for AI tools in imaging, the other three concerns do overlap with our survey findings where respondents expressed concern about the accuracy, security and trustworthiness of AI.

It is difficult to draw a direct comparison between adult and children’s views in the wider literature due to differences in questioning, nonetheless compared to the survey by Ongena et al [24] (upon which ours was based), adults on average more strongly agreed with the sentiment that it was important to get the scan results as fast as possible (score 4.49 in adults versus 3.89 for children (out of 5 on a Likert scale: 1 = strongly disagree, 3 = neutral, 5 = strongly agree)); scored similarly for worrying about data falling into the wrong hands (3.32 for adults versus 3.44 in children); and similarly for stating that even if AI was better at evaluating scans, they’d still prefer a doctor to review the study (3.32 for adults versus 3.51 in children). Regarding the fact that AI might replace doctors one day, adults scored an average of 3.50 versus 3.75 for children. The wish for faster results from the adult survey (compared to children) may reflect priorities in returning to work and life pressures (e.g. caring responsibilities). Other prior publications evaluating adult patient’s perceptions of AI in radiology have found similar thematic results to our survey of children and young adults. Most prefer human oversight of any AI tool and perceive any AI-based communication to lack emotional support, although they do welcome the use of AI if it can be proven to provide additional, accurate insights into their disease [24, 40, 41]. Other publications have additionally reported that a clear understanding of accountability and privacy concerns were a key factor in patient’s attitudes towards using AI-based healthcare solutions [18, 42, 43], including what using AI may mean with regard to clinical decision-making and access to healthcare professionals [44].

Comparing patient views (adults and children) with those of healthcare professionals on AI in imaging is challenging given the different focus of survey questions. Nonetheless, some similarities are noted—in one survey of healthcare professionals working in paediatric radiology [10], most agreed that their jobs were not at risk (85.4%), and that AI results should be still checked by a human (83.3% agreement). They also stated that diagnostic accuracy (32.1%), workflow efficiencies/speed (25.0%) and safety (22.5%) were the most important factors for consideration in AI design and implementation. In a different study of medical students [45], most (56%) were not convinced that AI could help with establishing definitive diagnoses in medicine and most agreed (83%) that AI would not replace radiologists. Finally, those working in mostly IT and industry were less trusting of AI, with only 25% stating they had confidence of AI results and 17% believing that the use of AI would mean healthcare staff could spend more time with patients, although they had high expectations of AI in the future with 86% believing medicine could become more efficient [46].

There are several limitations of our study. Our open recruitment strategy may have introduced a bias in the type of respondent that came forward to complete our survey. These are likely to have been from families and schools where access to digital devices and the internet were more accessible, those with English as a primary language and, potentially those from higher socioeconomic backgrounds (although we did not specifically ask about this detail). Our demographic representation may therefore not have included all possible ethnic backgrounds, although we received a large number of Asian participants, and our respondents did come from all four nations of the UK, indicating a broad reach of this survey.

We also focussed our survey mainly on opinions about AI for the diagnosis musculoskeletal disease on radiographs because we believed this was the most realistic clinical scenario for AI usage in the near future and a common disease many respondents would be familiar with. Whilst we did ask questions on more general areas, including cancer, brain and heart disease, we acknowledge that our findings may not generalise to all areas of paediatric imaging and more specific surveys on AI usage for those particular areas may be required. Furthermore, whilst we asked children to self-rate their computer literacy skills and awareness about AI, we do understand this is subjective, and there could be mixed perceptions about what ‘AI’ actually means. Nonetheless, from recent research conducted by a communications regulator in the UK (Ofcom) [47, 48], it was found that 59% of 7–17-year-old internet users have used a generative AI tool in the past year, with various international initiatives now promoting AI education programmes in school [49, 50], suggesting growing awareness and appreciation for this technology.

Our survey questions sought to strike a balance between being comprehensive, feasible but also understandable and not too tedious for children and young people to answer. To this end, we had to limit the number of questions we could ask, which we based these on the areas of priority guided by our GOSH PPIE Steering Committee. Future work could include smaller focus group or individual interviews with children and young people to delve deeper into some of the core issues surrounding accountability and ethical considerations to get more granular details on their opinions for these areas, in addition to exploring further their understanding of medical imaging tools and reasons for missed diagnoses when attending hospital. A survey exploring parental / carer viewpoints of the same questions may also be helpful in understanding if there is a difference between those of adults versus children, and whether there is further work needed to satisfy the needs of both (i.e. parent/carer and child) in the healthcare setting when considering AI implementation. Past studies reviewing caregiver and parental opinions on novel technologies in healthcare (e.g. robotics, virtual realtiy [38, 51, 52]) have generally shown high acceptance rates provided there are proven patient benefits with a careful consideration of possible risks of harm and how to mitigate these.

Finally, our survey was limited in our response rate due to limited uptake (although it still remains the largest survey of children and young persons views of AI for their medical imaging). Our survey, despite being based on an adult-validated survey of AI opinions for imaging and adapted with our PPIE steering group committee was not itself validated [53], and we had a wide range of ages of respondents with few non-adolescents. These included a single respondent aged 6 years old, two 12-year-old females, one 13-year-old male. Nevertheless, re-reviewing these individual concerns and replies to the survey questions did not deviate from those of the wider group.

In conclusion, children and young people in our survey population indicated that AI should be integrated into modern healthcare with an overwhelming preference for medical professional oversight for checks and balances. Our key messages from this survey should be considered by any hospital or radiology department looking to implement AI tools for children and young people so that their opinions and views are not forgotten. Further research into some aspects covered by our survey (e.g. ethical implications and accountability) from a wider population of respondents, or in depth subject-specific surveys, may be of benefit for future research.