Dissertations & Theses (Open Access)

Date of Award

Spring 5-2020

Degree Name

Doctor of Philosophy (PhD)


Momiao Xiong Phd

Second Advisor

Goo Jun Phd

Third Advisor

Suja S Rajan Mha Ms Phd


Personalized approaches have shown great potential to transform modern medicine. As challenging as it may sound, we are making tremendous progress with the help of data sciences and machine learning. Two fundamental tasks in data sciences are prediction and inference. In this dissertation, I proposed to address these two tasks using deep learning approaches in the setting of personalized medicine. First, I developed a novel framework to estimate individualized treatment effects (ITE), which quantified the variation in response to the same treatment among patients with heterogeneous profiles. The ITE estimation has the potential to replace the one-size-fits-all average treatment effects (ATE) commonly used in clinical practice and provides more accurate patient-specific treatment guidance. Second, I developed a statistical test to determine pairwise causation between two sets of continuous variables. Despite of the massive data available, the primary methods to determine causation clinically are through randomized controlled experiments and animal studies, which are highly inefficient or sometimes even infeasible. With this new statistical test, we were able to draw causal conclusions from observational data instead of experimental data alone, which was beneficial in terms of understanding underlying disease mechanisms. Statistical simulation was conducted to demonstrate the validity and accuracy of the proposed methods. Last but not least, I applied the developed methods on real-life datasets to demonstrate their usage. The TCGA lung cancer dataset was used to estimate ITE for patients with complex covariate structure. I also performed an end-to-end causal discovery for Alzheimer’s disease using the medical images from the ADNI dataset. The results indicate deep learning based approaches offer great flexibility and deep insights for biomedical data, which will help us bridge the gap in precision medicine.