Ethical issues on artificial intelligence in radiology: how is it reported in research articles? The current state and future directions

Abstract

Background. This paper evaluates the status of reporting information related to the usage and ethical issues of artificial intelligence (AI) procedures in clinical trial (CT) papers focussed on radiology issues as well as other (non-trial) original radiology articles (OA). Material and Methods. The evaluation was performed by three independent observers who were, respectively physicist, physician and computer scientist. The analysis was performed for two groups of publications, i.e., for CT and OA. Each group included 30 papers published from 2018 to 2020, published before guidelines proposed by Liu et al. (Nat Med. 2020; 26:1364-1374). The set of items used to catalogue and to verify the ethical status of the AI reporting was developed using the above-mentioned guidelines. Results. Most of the reviewed studies, clearly stated their use of AI methods and more importantly, almost all tried to address relevant clinical questions. Although in most of the studies, patient inclusion and exclusion criteria were presented, the widespread lack of rigorous descriptions of the study design apart from a detailed explanation of the AI approach itself is noticeable. Few of the chosen studies provided information about anonymization of data and the process of secure data sharing. Only a few studies explore the patterns of incorrect predictions by the proposed AI tools and their possible reasons. Conclusion. Results of review support idea of implementation of uniform guidelines for designing and reporting studies with use of AI tools. Such guidelines help to design robust, transparent and reproducible tools for use in real life.

Ähnliche Arbeiten