[EI] An Interactive XAI Interface with Application in Healthcare for Non-experts

Abstract

Explainable artificial intelligence (XAI) has gained increasing attention in the medical field, where understanding the reasons for predictions is crucial. In this paper we introduce an interactive and dynamic visual interface providing global, local and counterfactual explanations to end-users, with a use case in healthcare. The dataset used in the study is about predicting an individual’s coronary heart disease (CHD) within 10 years using the decision tree classification method. We evaluated our XAI system with 200 participants. Our results show that the participants reported an overall good assessment of the user interface, with non-expert users showing a higher satisfaction than users who have some degree of knoweldge of AI.

Paper PDF Link: https://research-information.bris.ac.uk/ws/portalfiles/portal/379782666/XAI2023Hu.pdf

Demo Link: med.bristol-xai.com

Demo Link (backup): med-xai.jingcs.com

*Counterfactual generation demonstrations are temporarily disabled in the interface due to server budget limits. Our implementation is based on DiCE, and here provides a detailed introduction and user-friendly examples of explanations generated by DiCE.

Ref

@inproceedings{hu2023interactive,
  title={An Interactive XAI Interface with Application in Healthcare for Non-experts},
  author={Hu, Jingyu and Liang, Yizhu and Zhao, Weiyu and McAreavey, Kevin and Liu, Weiru},
  booktitle={World Conference on Explainable Artificial Intelligence},
  pages={649--670},
  year={2023},
  organization={Springer}
}