Results show a wide variety of applications of AI that (will) influence the work of radiographers, ranging from changes in everyday workflow, like patient checks, planning of examinations, acquisition of images and post-processing activities, to changes in work flexibility, like cross-modality employability or performing radiologist tasks, and training, implementing and quality control of AI systems. Knowledge of AI, the basics as well as pitfalls, challenges, ethical and legal complications is prerequisite for radiographers. Three main points emerge from the scoping review, summarized in Figure 1 and described in more detail below.
(Potential) Applications of AI in the current workflow for radiographers.
In their review article, Hardy & Harvey provide an overview of where AI can play a role regarding the work (workflow) of radiodiagnostic technicians (3). AI can play a role in the entire process from the moment the patient arrives to processing of the data when the patient has returned home (Figure 2). Several other review articles also discuss the different tasks and steps within the radiological workflow in which AI can assist or even take over (5–7).
A similar overview in workflow is given for radiotherapy (8–12). Especially time-consuming, labour-intensive and repetitive steps can be taken over and standardized using AI to improve quality (accuracy and consistency) (11,12), such as automatic segmentation, dose calculation and treatment planning (Figure 2). Quality assurance (QA), an essential but also labour-intensive part of radiotherapy, can be automated with the help of AI (9,11). AI can also be applied in the entire workflow in nuclear medicine (13).
(Potential) task shifts by AI for radiographers.
Hardy & Harvey indicate in their review article that various applications of AI will lead to a change in roles and activities of radiographers (3). These include a greater role in interpersonal communication with the patient, making work more flexible (taking over tasks from the radiologist as well as employability at multiple departments) and a role in training, implementing and checking of AI systems. For this radiographers must not only be able to handle AI applications, they also must be able to critically assess the results of these AI applications (14). Here, data quality, ethical and legal aspects play a major role.
For AI the quality of input datasets is important for a correct outcome. The data is often trained and tested in a laboratory situation on a specific test set (14,15). This means that AI image recognition software must first be trained in each hospital on locally produced images. It is expected that radiographers will play an important role in this, but they are not yet prepared. Ethical and legal aspects also play a role here (14). It is not only about who is responsible if an AI application does something (wrong), but also about how you can share data (e.g. for training) within current legislation regarding personal data (15). Also, current regulations (European medical device regulations MDR 2017/745) require human supervision, evaluation and control of clinical automated systems (‘human oversight’) (3).
AI education and training for radiographers.
AI is developing at high speed, much faster than the education for professionals (i.e. radiographers) adapts to it. Therefore, there is a gap between AI applications that currently find their way into practice and knowledge about AI within that practice. This can lead to problems because the outcomes of AI cannot simply be trusted and are often not explainable, the functioning of an AI algorithm is seen as a black box (15,16). As a result, and because the average radiographer still has little affinity with AI, implementation of AI is lagging behind developments (17,18).
The current situation therefore strongly argues in favor of sound education in AI for radiographers and student radiographers (14,17,18). However, there is still little attention for AI in the curricula of radiographer courses at universities of applied sciences in the Netherlands. A national survey conducted among radiographers about their wishes and needs regarding information about or training on AI showed that the vast majority of respondents already encounter AI applications in their everyday work and feel a need for additional training (Figure 3).
Especially the basic principles, applications of AI and security are subjects radiographers want to gain more knowledge about. A majority prefer e-learning or a post vocational course, but a bachelor course or in-company training are also mentioned (19). Various courses on AI can be found online, but these are general and not focused on the specific field of the radiographer, although some are healthcare based (see Figure 4 for some examples).
Charow et al. made an overview of all healthcare-based AI educational programs that have been provided by institutions worldwide in the past and present (18). The article concludes that most programs focus on developing and implementing AI. It argues that education programs should be designed in such a way that healthcare professionals, in addition to safely applying AI, also learn to adapt and shift their tasks to remain relevant, and provides an example of the ideal AI curriculum (Figure 5). According to Chamunyonga et al. the challenge is to predict, based on developments in clinical workflow, what knowledge and skills current and future radiographers will need to perform their profession in a safe and competent way (20). According to this article, radiographers, in training to become experienced professionals, must have an understanding of terminology, underlying concepts, applications within the field and the potential added value therein. Furthermore, suggestions for the curriculum are given: knowledge of limitations and risks, ethics and legislation, quality and safety, multidisciplinary care and research.