CHERIE: User-Centred Development of an XAI System for Chest Radiology through Co-Design

Abstract

Modern medical imaging systems increasingly use AI for analysis and diagnosis, yet the "black-box" nature of current deep learning algorithms limits their practical use in radiology. Explainable AI (XAI) aims to address this by making AI decisions more transparent and interpretable. In medical imaging, XAI tools often highlight critical regions in images to explain AI decisions, but their complex visual explanations and poor UI design impede their clinical adoption. This study introduced CHERIE, an XAI prototype designed to enhance transparency in AI-assisted chest radiology. Using our pre-developed XAI diagnostic tool for chest radiology, we adopted a user-centered design (UCD) methodology to develop user interfaces for the AI-enabled diagnostic tool. In particular, we engaged medical practitioners, AI developers, and HCI experts in a multidisciplinary co-design workshop. This collaborative effort was crucial in identifying requirements from the user perspectives, aiming to boost understanding and trust in AI-driven diagnostics. Our findings emphasise the need for UCD for the adoption of XAI systems, proposing user requirements to seamlessly integrate these systems into clinical workflows and effectively address end-user needs.