Categorising challenges and solutions towards ethical AI in breast cancer treatment: a rapid umbrella review complemented by participatory methods

Abstract

Abstract Ethical issues in Artificial Intelligence (AI) are central to the global political and scientific agenda. However, existing guidelines and regulations are generic across distinct AI applications in societal domains. Analysing the use of AI for breast cancer treatment as a case study, this work aims to review the main biases brought up by AI in healthcare and set the grounds for categorising such issues. We combined a literature review with participatory research methods to investigate this emerging topic. These consisted of reviewing the state-of-the-art, through a rapid umbrella review, complemented by stakeholder consultation in social innovation sessions, and interviews. These results were combined and analysed through Rapid Qualitative Analysis. Our results clearly show that challenges are multicomplex and need to be structured into complementary streams, leading to the following categorisation of ethical challenges: (1) Individual (human) challenges, such as the lack of adequate training, individual beliefs and pre-conceptions; (2) Technical challenges, for example, poor algorithm design or skewed training datasets; (3) Organisational challenges, e.g., lack of diversity in teams or lack of audit methods; and (4) Societal challenges, such as health inequities, discrimination or lack of adequate regulations. Several practical examples fitting each of these areas and potential mitigation measures are described, as well as areas for future research. Consequently, a robust ethics-by-design framework informed by broad multistakeholder engagement as demonstrated through our participatory methods, is essential for anticipating and mitigating bias in AI healthcare and promoting a fairer use of AI in health.

Ähnliche Arbeiten