Shef.AI / Turing IG seminar: Learn to Impute and Align with Knowledge: Striving for Unbiased and Clinically Accurate AI in Healthcare by William Cheung.

This seminar is jointly organised by the Centre for Machine Intelligence at the University of Sheffield and Alan Turing Institute Interest Group on Meta-learning for Multimodal Data.

Abstract:
AI has been rigorously explored in healthcare in recent years. Most of the AI models for healthcare are learned from the EHR data. To achieve unbiased and clinically accurate AI, there are specific issues which should be carefully addressed. One key challenge is how to mitigate bias in EHR predictive analytics caused by data missingness. In addition, how to properly align the inferred models with prior clinical knowledge is another, which is particularly important in healthcare. In this talk, I will present some of our recent work to address these two key challenges. The first part of the talk will be focusing on data analytics methods for joint imputation and prediction to address missingness in binary EHRs and multi-modal EHRs. The second part will be on automatic radiology report generation and retrieval based on vision-language models. We propose methodologies to guide the model learning by prior medical knowledge to enhance the clinical accuracy for the report generation. The datasets used in our work for benchmarking include MIMIC, eICU, IU XRay, and MIMIC-CXR.

Biography:
William K. Cheung received the Ph.D. degree in computer science from the Hong Kong University of Science and Technology in 1999. He is currently the Associate Vice President (Undergraduate Programmes) and Professor of Computer Science at Hong Kong Baptist University. His research interests include artificial intelligence, data mining, social network analysis, and healthcare informatics. He has served as the PC members of AAAI, IJCAI, NeurIPS, UAI, ICLR, etc., and the programme co-chairs for a number of international conferences and workshops on areas including artificial intelligence, machine learning, data mining, Web intelligence, and health informatics. From 2002-2018, he was on the Editorial Board of the IEEE Intelligent Informatics Bulletin. He is a Track Editor of Web Intelligence Journal and an Associate Editor of Journal of Health Information Research, and Network Modeling Analysis in Health Informatics and Bioinformatics. In recent years, he is also heavily involved in implementing transdisciplinary undergraduate programmes in the areas of health innovation, arts and technology, and digital humanity.