Variation in standards for reporting artificial intelligence research using burn image datasets raises ethical questions: A scoping review.

Abstract

Accurate assessment of burn depth and area are required to guide treatment and inform prognosis. Currently this assessment relies on subjective visual analysis by clinicians. A sub-field of artificial intelligence (AI), computer vision, has been identified as a tool that could bring objectivity to this analysis. This has resulted in a fast-growing area of interdisciplinary research. Given that the regulation of AI research in healthcare is in its infancy, this scoping review aimed to identify current practice and if further guidance for researchers is required. A scoping review of computer vision studies in burn care was conducted using PubMed, arXiv and IEE Xplore databases to ascertain current standards of practice of AI research involving burn image datasets. A quantitative and qualitative analysis and synthesis of papers was undertaken and themes were identified. The database searches identified 1562 papers, of which 75 met the review inclusion criteria. There was variation in the standards of reporting in the reviewed papers. Key common ethical issues identified included: unavailable code and data, missing statements about consent and ethical review, and paucity of information about the datasets. Omissions in the reporting of many studies may raise ethical questions or reflect the need for standardised reporting. A preliminary checklist tailored to the reporting of burn image datasets that are used in AI research is suggested to support researchers working in this important interdisciplinary field.