FAIR human neuroscientific data sharing to advance AI driven research and applications: Legal frameworks and missing metadata standards

Abstract

Modern AI supported research holds many promises for basic and applied science. However, the application of AI methods is often limited because most labs cannot, on their own, acquire large and diverse datasets, which are best for training these methods. Data sharing and open science initiatives promise some relief to the problem, but only if the data are provided in a usable way. The FAIR principles state very general requirements for useful data sharing: they should be findable, accessible, interoperable, and reusable. This article will focus on two challenges to implement the FAIR framework for human neuroscience data. On the one hand, human data can fall under special legal protection. The legal frameworks regulating how and what data can be openly shared differ greatly across countries which can complicate data sharing or even discourage researchers from doing so. Moreover, openly accessible data require standardization of data and metadata organization and annotation in order to become interpretable and useful. This article briefly introduces open neuroscience initiatives that support the implementation of the FAIR principles. It then reviews legal frameworks, their consequences for accessibility of human neuroscientific data and some ethical implications. We hope this comparison of legal jurisdictions helps to elucidate that some alleged obstacles for data sharing only require an adaptation of procedures but help to protect the privacy of our most generous donors to research … our study participants. Finally, it elaborates on the problem of missing standards for metadata annotation and introduces initiatives that aim at developing tools to make neuroscientific data acquisition and analysis pipelines FAIR by design. While the paper focuses on making human neuroscience data useful for data-intensive AI the general considerations hold for other fields where large amounts of openly available human data would be helpful.

From Lab to Life: Challenges and Perspectives of fNIRS for Hemodynamic-based Neurofeedback in Real-World Environments

“…The advancement of the neurofeedback field and most of the ideas discussed depends, among other things, on a common factor: the transparent and open exchange of data and analysis codes. Especially when it comes to AI applications and machine learning algorithms, access to large data sets is required to train and validate models and to reduce bias [190,191]. It is likely that AI will play a role for neurofeedback in the future [192], so the availability of diverse and large data sets will be crucial in the development of AI models.…”