Developing a Metrological framework for Assessment of Image-Based Artificial Intelligence systems for disease detection

基于图像的人工智能(AI)系统正在日益发展，这些工具在不同的临床环境中是健壮和有效的，这是至关重要的。迄今为止，由于没有核准的评价准则，对业绩进行了临时评估。大多数研究都有方法上的弱点和不可比拟的结果。一个标准化和公正的人工智能工具性能、通用性和适宜性评估框架将满足这些需求，并使基于图像的人工智能系统能够更有效、更可靠和更可重复地验证疾病检测。这个项目将以乳腺癌筛查为样本，为这样一个框架的设计提供信息。

Image-based artificial intelligence (AI) systems for disease detection are increasingly being developed, and it is vital that these tools are robust and effective in heterogeneous clinical settings. To date, performance has been assessed in an ad hoc manner as there are no approved guidelines for evaluation. Most studies have methodological weaknesses and results that are not comparable. A standardised and impartial framework for performance, generalisability, and suitability assessment of AI tools will address these needs and enable more efficient, reliable, and reproducible validation of image-based AI systems for disease detection. This project will use breast cancer screening as the exemplar to inform the design of such a framework.