The FUTURE-AI project presents an international consensus guideline aimed at ensuring trustworthy and deployable AI throughout its entire lifecycle in healthcare settings.
Authors: Stijn Van de Maele, Nikolaos Dikaios, Jacob Jaremko, Geert Litjens, Marc Dewey, Marc Niethammer, Carolina Wählby, Rickmer Braren, Mattias P. Heinrich, Rick G. Bartram, Ivana Isgum, Nikita Moriakov, Andrew F. Frangi, Wiro J Niessen, and Jens Kleesiek.
Published in: The BMJ, 2024;388:e081554. https://doi.org/10.1136/bmj-2024-081554
Key Components (FUTURE-AI Acronym)
The structure of the guideline revolves around six crucial domains:
- Foundation: Establishing core principles such as fairness, transparency, and accountability.
- User-Centered Design: Ensuring AI tools are developed with genuine input and evaluation from patients and clinicians.
- Traceability: Maintaining clear version control, logging, and audit trails for AI processes.
- Usability: Crafting intuitive interfaces and workflows that integrate smoothly into clinical practice.
- Reliability: Rigorous testing to confirm robustness, safety, and performance consistency.
- Explainability: Providing understandable reasoning behind AI-driven decisions for users and stakeholders.
Methodology
Guideline development followed a modified Delphi process, with multiple rounds of expert consultation (clinicians, AI developers, ethicists, regulators) to reach consensus on best practices .
Significance
By offering detailed, lifecycle-spanning recommendations, FUTURE-AI addresses practical and ethical concerns tied to medical AI adoption—namely avoiding bias, ensuring patient safety, and fostering trust among users bmj.com+9bmj.com+9bmj.com+9.
Implications for Healthcare Stakeholders
- Developers gain structured guidance for designing and documenting AI systems.
- Healthcare providers are equipped to evaluate the readiness and trustworthiness of AI tools.
- Regulators receive a benchmark for policy-making and certification processes aimed at safe, ethically grounded AI deployment.
READ THE FULL ARTICLE ON BMJ