Metrology for artificial intelligence in medicine

Abstract

To build a metrological basis for underpinning the certification and validation of Artificial Intelligence (AI) algorithms and AI-based measurement devices in medicine, the programme Metrology for Artificial Intelligence in Medicine (M4AIM) was started at the German National Metrology Institute PTB in 2021 as one of the pilot projects of the German digital quality infrastructure (QI-Digital) initiative. The aim of the programme is to develop standard metrics for the explainability, robustness, and uncertainty of AI performance and data quality assessment. M4AIM comprises fourteen projects addressing different topics in several fields of the medical use of AI methods, with methodological approaches ranging from fundamentals of machine learning to its use in specific applications. Generating trustworthy synthetic reference data is a common topic among many of the projects. This article gives an overview of the approaches explored, highlights some of the results obtained and discusses the interrelation of metrology and machine learning from a more general perspective.

Ähnliche Arbeiten