Trust perceptions towards XAI in healthcare

• Main Author: Cai, Xinrui
• Other Authors: Fan Xiuyi
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2024
• Subjects: Computer and Information Science
• Online Access: https://hdl.handle.net/10356/175388
• Institution: Nanyang Technological University

This study explores the perceptions of medical professionals towards artificial intelligence (AI) and the influence of Explainable AI (XAI) algorithms on their trust in AI. The research focuses on one-on-one interviews conducted with 12 medical students, divided into two groups of six, each group exposed to one of two XAI algorithms: SHAP (SHapley Additive exPlanations) and LIME (Local Interpretable Model-agnostic Explanations), explaining how Random Forest Classifier predicted lung cancer from a set of predictive attributes including demographic information, basic health metrics, lifestyle habits and findings from genetic PCA (Principal Component Analyses). Additionally, online surveys were administered to 50 medical students on the same case as interviews. Both interviews and surveys had selected participants ensuring equal representation from two medical schools in Singapore, with considerations for gender and self-rated confidence in AI knowledge. Qualitative data from interviews were analysed using Reflexive Thematic Analysis, revealing themes related to trust in AI and perceptions of XAI algorithms. Quantitative data were analysed using Microsoft Excel to visualize trends and patterns in the survey responses. Four research questions were developed in this study and the findings suggest that the type of XAI algorithm used does not significant impact medical professionals' trust in AI, and XAI’s impact on trust of medical professionals on AI may not be direct and requires more research and improving on XAI output to achieve its intended purposes. This study contributes to the understanding of how XAI can enhance trust in AI among medical professionals, with implications for the design and implementation of AI systems in healthcare settings.