[ad_1]
Discussion
In this study, we have investigated Swedish breast radiologists’ views on the use of AI in mammography screening. The respondents were, to a large extent, positive towards the integration of AI in screen reading, especially those having difficulties finding the time to perform screen-reading. This could explain the slightly more positive attitude, compared with general studies on radiologists’ attitudes towards AI.23 26 27 We could not establish a correlation between attitude and AI-literacy, prevalent in previous general studies.23 However, it needs to be taken into account that our sample represents a relatively small number of individuals. Those more opinionated about AI could also be more inclined to answer the survey, possibly inducing bias in the results. The specific context, of mammography screening and the profession of breast radiologists, in a digitally advanced welfare state, however, showcases the importance of considering technological implementations in relation to organisational and socioeconomic structures.
Furthermore, we did identify important reservations, factors associated with high uncertainties, and diverse viewpoints, such as regarding liability of AI use. The question is whether established practices need to be adjusted when medical decisions are increasingly supported by automated technologies or AI-systems. Our results point to a somewhat higher perceived responsibility of radiologists in AI-supported radiological practice, compared with previous studies.20 27 Furthermore, the results show the complexity of accountability when AI enters radiology, how it is contextual, dependent on how AI is used and which actors are included. This further emerges in the insecurities regarding liability for missed cancers, when AI is used as a co-reader or as stand-alone reader, or when radiologists disregard AI findings. The results indicate a perceived shift of responsibility away from the radiologist as automation increases. Additionally, uncertainties regarding the responsibilities of AI-developers (and AI-systems) suggest a need for clarification.28
We could not identify one unified vision of a preferred way to use AI in mammography screening. Previously, AI has been expected to be used as second reader and for optimising workflows.16–18 While using AI as replacement for one reader in double-reading was the most preferred option in our study, many favoured using it as an addition to double-reading or in a combination of uses, suggesting perceived qualities other than workload reduction. Furthermore, the perceived risk of AI deteriorating working conditions might be due to several reasons. Besides a risk of eroded knowledge of the normal breast, reduced screen reading workload might not improve working conditions. While more time for patient-centred care is portrayed as a positive outcome of AI,27 some perceived screen reading as a welcomed interruption from emotionally burdensome work, which might be lost due to automation. Working with AI-systems also adds layers of interpretation,29 which could be exhausting. This seems to be perceived as a medical risk, but also as an ethical burden with legal uncertainties.
Additionally, AI in mammography screening needs to be considered in light of previous innovations. Some aspects are not unique for AI, such as contested expertise.30 However, radiologists, trainees and medical students strongly expect AI to change the field and impact job opportunities, tasks and relationships with patients.16–18 20 Our study shows that breast radiologists believed that AI will impact the profession, both positively and negatively. However, most did not believe it would impact the role of the breast radiologist. Few thought it would improve recruitment, possibly due to the idea of AI negatively affecting the professional reputation.26 Many considered, or were uncertain whether, implementing AI represents a substantial difference in comparison to previous technologies. While new imaging methods aim to improve cancer visibility, AI differs since it involves medical decision-making. This implies that social, ethical and legal aspects have to be addressed, which in turn depends on how AI is incorporated into the clinical workflow. Greater unclarity about accountability seems to be prevalent regarding AI as a stand-alone technique, which was also the least favoured approach. This suggests that physicians are not willing to renounce their responsibility in medical decision-making. In total, our results echo the need for more research on social, ethical and legal matters of integrating AI into radiology and screening.
Strengths and limitations
The main limitations of the study are the specific conditions of the Swedish setting and the small number of respondents. The response rate was satisfactory, but the target population was limited since there are few Swedish breast radiologists. The study’s strengths were that the respondents were highly experienced in breast imaging and that half of the group had experience of using AI in breast imaging.
[ad_2]
Source link
