FUTURE-AI: Guiding Principles and Consensus Recommendations for Trustworthy Artificial Intelligence in Medical Imaging

Abstract

The recent advancements in artificial intelligence (AI) combined with the extensive amount of data generated by today's clinical systems, has led to the development of imaging AI solutions across the whole value chain of medical imaging, including image reconstruction, medical image segmentation, image-based diagnosis and treatment planning. Notwithstanding the successes and future potential of AI in medical imaging, many stakeholders are concerned of the potential risks and ethical implications of imaging AI solutions, which are perceived as complex, opaque, and difficult to comprehend, utilise, and trust in critical clinical applications. Despite these concerns and risks, there are currently no concrete guidelines and best practices for guiding future AI developments in medical imaging towards increased trust, safety and adoption. To bridge this gap, this paper introduces a careful selection of guiding principles drawn from the accumulated experiences, consensus, and best practices from five large European projects on AI in Health Imaging. These guiding principles are named FUTURE-AI and its building blocks consist of (i) Fairness, (ii) Universality, (iii) Traceability, (iv) Usability, (v) Robustness and (vi) Explainability. In a step-by-step approach, these guidelines are further translated into a framework of concrete recommendations for specifying, developing, evaluating, and deploying technically, clinically and ethically trustworthy AI solutions into clinical practice.

Ethics of AI in Radiology: A Review of Ethical and Societal Implications

“…Adopting an intersectional perspective that takes into consideration how different traits of our identity intersect will be crucial, especially in the case of breast cancer. As previous research has shown, other factors that intersect with gender contribute to the formation of bias, such as ethnicity, skin color, socioeconomics, geography or breast density (Lekadir et al, 2021). In this regard, the issue of gender bias in female-only datasets requires a more detailed analysis.…”

“…These questions are accompanied by concerns that AI systems could perpetuate or even amplify ethical and societal injustices. Based on key ethical values such as respect, autonomy, beneficence, and justice (Beauchamp and Childress, 2001), several guiding principles and recommendations have been formulated to tackle these issues (Currie et al, 2020;Ryan and Stahl, 2020)-Such principles and recommendations have also been communicated on EU level (High-Level Expert Group on Artificial Intelligence, 2019), and initiatives such as FUTURE-AI (Lekadir et al, 2021) have been started, which have been developed to ensure that advances in AI systems and advances in AI ethics do not contradict one another.…”

Sharing Generative Models Instead of Private Data: A Simulation Study on Mammography Patch Classification

“…To further validate the benefits of inter-centre generative model sharing, we recommend future work to test prediction performance improvement on full mammograms apart from patches and on further datasets across (imaging and non-imaging) modalities, organs, and standardised clinical tasks [18] with universal classification objectives (e.g., benign/malignant apart from healthy/non-healthy classification). We further propose future work to explore the effect of initialisation of classification models with pretrained weights alongside the effect of traditional preprocessing and data augmentation techniques, such as histogram and intensity scale normalization techniques [11].…”

“…Scarcity of expert-annotated medical images often constrains deep learning based methods to be trained and evaluated on a small dataset coming from a single centre [8]. Accordingly, such methods suffer from lack of generalisation and robustness [18]. A solution to this problem is inter-centre data sharing.…”

Clinician's guide to trustworthy and responsible artificial intelligence in cardiovascular imaging

“…Summary of potential questions to support the discussion of AI trustworthiness between clinicians and technical experts adapted from Ammanath (8) and Lekadir et al (55).…”