Authors' objectives: The integration of artificial intelligence in medical image classification for screening has the potential to enhance efficiency, diagnostic accuracy and accessibility. However, ethical concerns such as accountability, bias, transparency and the impact on healthcare professionals remain critical. This review synthesises qualitative evidence on the ethical considerations surrounding artificial intelligence adoption in screening programmes. Artificial intelligence (AI) enables machines to mimic human abilities like learning and decision-making. Machine learning, especially deep learning, allows AI to identify patterns in data without explicit programming. Integrating AI into health care has the potential to enhance medical imaging and diagnostic accuracy, as AI-based technologies can assist in detecting medical conditions such as breast cancer and diabetic eye diseases. However, its lack of transparency and reliance on unsupervised learning raise ethical concerns that need careful consideration. Breast cancer is the most common cancer in UK women. Early detection through screening, such as the NHS Breast Screening Programme for women aged 50–70 years, can lead to better outcomes. Breast cancer screening generates vast amounts of data, making it ideal for training AI algorithms. Screening is costly and time-intensive, requiring expert radiologists to interpret complex images. In the UK, mammograms are assessed by two specialists, with arbitration for disagreements. Within this framework, AI has the potential to improve efficiency, reduce costs and streamline the screening process by assisting or replacing human radiologists and helping make decisions on whether further assessment is needed. Diabetes affects over 5.6 million people in the UK and can lead to diabetic retinopathy, a major cause of blindness. The UK Diabetic Eye Screening Programme (DESP) invites diabetic patients aged ≥ 12 years for regular screenings, where eye images are assessed by trained graders in multiple stages. With rising diabetes cases increasing the demand for specialists, AI has the potential to assist in image analysis, improving efficiency and reducing costs, similar to its role in breast cancer screening. Recent advancements in medical AI have focused on breast cancer and diabetic retinopathy screening. Programmes/initiatives, such as Digital Mammography DREAM Challenge, the Food and Drug Administration-approved EyeArt system, collaborations involving Google DeepMind, NHS Trusts and universities, are achieving accuracy in disease detection nearing expert radiologists’ levels and further refining AI in screening. Artificial intelligence has the potential to revolutionise medical screening and diagnosis, particularly for breast cancer and diabetic retinopathy, by enhancing accuracy and minimising human errors. As AI becomes more integrated into medical imaging and classification, ethical, legal and social concerns have been widely discussed, balancing enthusiasm over its benefits with caution about potential risks. The UK National Screening Committee (NSC)’s criteria, for evaluating the viability, effectiveness and appropriateness of a screening programme, incorporate the ethical principles of health screening. The ethics framework recently developed by UK NSC includes the four following principles aiming to guide the goals of screening: improve health and well-being, treat people with respect, promote equality and inclusion, and use public resources fairly and proportionately. There are currently no specific ethical frameworks for AI in screening, but general healthcare AI guidelines can be adapted for screening. World Health Organization guidance outlines six principles for ethical AI in health: protecting autonomy, promoting well-being, ensuring transparency, fostering accountability, ensuring equity and supporting sustainability. In addition, The Alan Turing Institute provides guidance for the ethical and responsible design of AI systems in the public sector. It emphasises integrating ethical principles to ensure fair, safe and reliable systems while addressing potential harms like bias and privacy invasion. The framework includes values, like respect, connect, care and protect, and principles of fairness, accountability, sustainability and transparency. It stresses governance, accountability, transparency and proactive risk management to foster trust and prevent harm. The aim of the current review is to identify and synthesise qualitative evidence on ethical issues related to the application of AI-based technologies for medical image classification in screening to inform stakeholders about the advancement of ethically responsible innovation in AI. The review aimed to answer the following question: What are the ethical implications of the use of AI-based technologies for medical image classification systems in screening?

Authors' results and conclusions: Fourteen qualitative studies were included, covering perspectives from clinicians, radiologists, artificial intelligence developers, policy-makers and patients. Key ethical concerns identified included: (1) the necessity of human oversight to ensure that artificial intelligences diagnostic recommendations are appropriate; (2) challenges in assigning liability when artificial intelligence errors occur; (3) risks of algorithmic bias due to discrepancies between training data sets and real-world populations; (4) concerns over data privacy, cybersecurity and informed consent in artificial intelligence-driven decision-making; (5) the need for transparency in artificial intelligence decision-making processes to build trust and (6) potential deskilling of healthcare professionals and shifts in professional responsibilities. While artificial intelligence was seen as a valuable tool to augment clinical decision-making, stakeholders emphasised that ethical frameworks must guide its implementation to maintain public trust and patient safety. This review highlights the critical considerations that must be addressed to ensure the responsible integration of artificial intelligence in medical screening. Policy-makers, healthcare institutions and developers should prioritise human oversight, robust regulatory frameworks and strategies to mitigate bias and ensure transparency. Future research should focus on disease-specific artificial intelligence applications and long-term ethical implications. Fourteen primary qualitative studies were judged to be relevant to the review question and included in the review. Studies were relevant to different target conditions, including: breast cancer (n = 5), lung cancer/pulmonary nodules (n = 2, one of which was also relevant to diabetic retinopathy), diabetic retinopathy (n = 2, one of which was also relevant to pulmonary nodules), prostate cancer (n = 2), cutaneous melanoma (n = 1), pulmonary hypertension (PH) (n = 1) or no specific target condition (n = 2). While these varied in clinical focus, common ethical issues, concerns and considerations related to the use of AI-based technologies for medical image classification in screening emerged across conditions. These included the importance of maintaining the ‘human in the loop’, liability/accountability, algorithmic bias and the impact of discrepancies between real-world and training data, the issues of data confidentiality, cybersecurity and consent, the issue of transparency, potential risks to doctors’ competencies and impact on the medical profession, the trade-off between false positives and disease detection, the issue of having lower tolerance for errors made by AI versus humans, potential cases of reduced effectiveness of AI, lack of familiarity and trust as a potential barrier to implementation, and additional requirements for implementation. Studies included patients (n = 6), healthcare professionals such as radiologists, primary care physicians, ophthalmologists (n = 5) or a range of professionals and stakeholder involved in the development and implementation of AI in radiology settings such as clinicians, radiologists, software developers or vendors (n = 5). Studies on patients included women of breast cancer screening age (n = 3), men with experience of prostate cancer screening (n = 1) and patients with PH (n = 1). Studies on healthcare professionals included breast cancer screening specialists (n = 3), ophthalmologists (n = 1) or radiologists with experience in prostate cancer screening (n = 1). The studies were conducted in the UK (n = 4; one of which was in Scotland), Sweden (n = 3), Australia (n = 2), Germany (n = 2) and one each in Norway, China and the USA. The studies examined a broad range of aims, including to understand whether AI and healthcare professionals see algorithmic bias in health care as a problem, to explore the early experiences of implementing and using an AI-based diagnostic decision support system in chest radiology, to explore the barriers and facilitators of the implementation of AI-supported devices for screening of diabetic retinopathy within general practitioner practice, and to explore perceptions and attitudes towards the use AI in breast cancer screening among women participating in a national breast cancer screening programme. Studies used a range of qualitative methodologies, most frequently semistructured interviews (n = 9). Other studies used focus groups (n = 2) or a combination of focus groups and semistructured interviews (n = 2) and a dialogue group methodology (n = 1). There was a wealth of qualitative evidence highlighting a range of ethical implications and considerations associated with the use of AI-based technologies for medical image classification systems in screening. The current review can inform stakeholders about the advancement of ethically responsible innovation in AI and issues to be considered before the application of AI in screening programmes in the UK. In particular, overarching concerns around algorithmic bias, transparency, accountability, data confidentiality and trust emerged as central themes that can directly impact implementation. The following methodological limitations of the review were identified: only EMBASE, Ovid MEDLINE, PsycInfo and CINAHL electronic databases were searched; searches were limited to records published from June 2020 onwards (updating the search conducted for the UK NSC 2021 evidence map on AI in DESPs, which included papers published since 2000) and included only peer-reviewed, English-language journal articles; and only 20% of the titles and abstracts were double screened. The volume of published evidence identified, although sufficient (based on prespecified criteria) to only include qualitative studies and exclude survey and questionnaire studies, was insufficient to enable the review to focus on the target conditions of breast cancer and diabetic retinopathy. As prespecified in the review protocol, the 20% validation was also applied in the quality assurance of data extraction and thematic analysis due to time constraints. A future review could potentially aim for double reviewing 100% of all steps of the review process to ensure accuracy and consistency. Within this framework, another potential limitation that may need consideration is the subjective nature of qualitative thematic analysis, which may not allow for the replication of findings despite double reviewing. As evidence continues to be produced, future reviews may focus on specific conditions like breast cancer or diabetic retinopathy to generate targeted insights on the ethical challenges, patient perspectives and implementation considerations of AI in screening of those conditions. Additionally, while existing studies provide relevant information, they were not designed to examine AI’s ethical implications in screening. Future qualitative research aligned with this focus could offer deeper ethical insights.

Authors' methods: A systematic search of qualitative studies, from June 2020 to September 2024, was conducted across multiple databases: MEDLINE, EMBASE, PsycInfo® (American Psychological Association, Washington, DC, USA) and Cumulative Index to Nursing and Allied Health Literature. Primary qualitative studies exploring healthcare professionals’, patients’ and other stakeholders’ perspectives on artificial intelligence in screening were included. Thematic analysis was performed, and findings were assessed using the Grading of Recommendations Assessment, Development and Evaluation-Confidence in the Evidence from Reviews of Qualitative Research approach to evaluate confidence in the evidence. Using a qualitative review methodology, all primary qualitative studies evaluating the views of health professionals, providers and users (clinicians and patients) of screening programmes and the general public on the ethical impact of AI-based technology in medical screening programmes or similar settings were included; if insufficient qualitative studies were identified, the review would have been extended to include questionnaires and/or surveys, but this was not necessary. Studies relevant to the use of AI in screening for breast cancer and diabetic retinopathy were initially prioritised for full-text assessment and inclusion. However, only a few relevant papers were identified and thus studies relevant to any medical condition were included. Studies were identified using database searches and were screened for relevance based on pre-defined inclusion and exclusion criteria. The literature search covered Ovid MEDLINE, EMBASE, PsycInfo® (American Psychological Association, Washington, DC, USA) and Cumulative Index to Nursing and Allied Health Literature (CINAHL), ensuring broad coverage of peer-reviewed studies. The searches conducted for Question 4 of the UK NSC 2021 evidence map were updated. Thus, searches were limited to the period from the date of the previous search (30 June 2020) to the date of the current searches (18 September 2024). Some additional search terms were included to broaden the original search. Included studies of identified systematic reviews were also assessed for inclusion. Studies were included if they met the following criteria: population: health professionals, patients and stakeholders involved in screening programs intervention: AI-based medical image classification technologies used for screening outcome: ethical concerns and perspectives related to AI use in screening study design: qualitative studies, including interviews and focus groups language and time frame: English-language studies published after 2000. Conference abstracts, non-peer-reviewed articles and non-English publications were excluded. Data extraction was conducted using a standardised form that collected information on study characteristics methodology and key themes. The Critical Appraisal Skills Programme checklist was used to assess the quality and potential bias of each included study. Thematic analysis was used to synthesise findings across studies. Key themes were identified by examining common patterns in participants’ views. The Grading of Recommendations Assessment, Development and Evaluation-Confidence in the Evidence from Reviews of Qualitative Research approach was used to evaluate the level of confidence in each finding. This assessment considered four factors: methodological limitations (quality of the included studies) coherence (consistency of findings across studies) relevance (applicability of the findings to AI in screening) adequacy (the richness and quantity of data supporting each theme). Findings were categorised as high, moderate, low or very low confidence, providing an indication of how strongly the evidence supports each ethical concern.

Project Status: Completed

URL for project: https://www.journalslibrary.nihr.ac.uk/programmes/hta/NIHR172233

Year Published: 2026

URL for published report: https://www.journalslibrary.nihr.ac.uk/hta/GJSE4912

URL for additional information: English

English language abstract: An English language summary is available

Publication Type: Full HTA

Country: England, United Kingdom

DOI: 10.3310/GJSE4912

Organisation Name: NIHR Health Technology Assessment programme

Contact Address: NIHR Journals Library, National Institute for Health and Care Research, Evaluation, Trials and Studies Coordinating Centre, Alpha House, University of Southampton Science Park, Southampton SO16 7NS, UK

Contact Name: journals.library@nihr.ac.uk

Contact Email: journals.library@nihr.ac.uk