Human-Curated Validation of Machine Learning Algorithms for Health Data

Abstract

Abstract Validation of machine learning algorithms that take health data as input is analysed, leveraging on an example from radiology. A 2-year study of AI use in a university hospital and a connected medical university indicated what was often forgotten by human decision makers in the clinic and by medical researchers. A nine-item laundry list that does not require machine learning expertise to use resulted. The list items guide stakeholders toward complete validation processes and clinical routines for bias-aware, sound, energy-aware and efficient data-driven reasoning for health. The list can also prove useful to machine learning developers, as a list of minimal requirements for successful implementation in the clinic.