Ethical guidelines for deploying artificial intelligence applications in the pathology field: Lessons learned from a...

Several artificial intelligence (AI) algorithms have been developed with inherent age, sex, gender, racial, and ethnic biases. In pathology, this leads to marked performance disparities across different demographic groups. In this article, we highlight the root of differences in representation in AI, list the probable causes and clinical implications of these gaps, and propose an ethical framework for addressing representation and bias in AI in pathology. Various studies have highlighted efforts to mitigate the gaps. However, to our knowledge, there are no standard guidelines in the field of pathology that ensure the fair use of AI to counter biases in representation. We propose a heuristic framework that is tailored specifically to lab medicine and pathology workflow. Based on data life cycle and pathology workflow, these guidelines are stratified into governance and leadership strategies, preprocessing phase standards, processing phase standards, operational deployment, and ongoing monitoring guidelines.