Deep learning, a proliferating subfield of machine learning, has shown success in various domains such as computer vision and natural language processing, where conventional statistical methods struggle with scalability and precision challenges. Its application can potentially impact many areas in healthcare ranging from automated diagnosis to healthcare policy making. In this talk, he will present two projects that he worked on in collaboration with Michigan Integrated Center for Health Analytics and Medical Prediction (MiCHAMP): (1) Using the image recognition techniques in deep learning, we built an automated grading system of endoscopic disease severity in ulcerative colitis. The system showed competitive performance in comparison with human experts on two independent testing sets. (2) We utilized the concept and algorithm of word embedding in natural language processing to build up a new framework of modeling the medical codes that are widely available in claim data, and show the efficacy in improving the prediction of unplanned readmissions.
Wenshuo Liu was a post-doctoral research fellow in the Department of Biostatistics & Medical Informatics and Department of Physics at University of Wisconsin-Madison from 2014-2017. He obtained his PhD in the field of theoretical physics from Rutgers University in 2014. His research focuses on methodology development in machine learning, especially deep learning for large-scale, complex data, with application in various areas, including medical imaging, health services research and multiple imputation of missing data.