Off-campus WSU users: To download campus access dissertations, please use the following link to log into our proxy server with your WSU access ID and password, then click the "Off-campus Download" button below.
Non-WSU users: Please talk to your librarian about requesting this dissertation through interlibrary loan.
Enhancing Healthcare Informatics Through Deep Learning With Graph-Based Models And Self-Distillation
Access Type
WSU Access
Date of Award
January 2024
Degree Type
Dissertation
Degree Name
Ph.D.
Department
Computer Science
First Advisor
Ming Dong
Abstract
Deep learning has revolutionized the areas of computer vision, natural language processing, robotics, medical image processing and many others. In this dissertation, we focus on enhancing healthcare informatics by modeling healthcare data with deep neural networks for different downstream tasks such as classification, regression and image segmentation. In Chapter 2 of this dissertation, we present the relevant deep learning literature based on which we proposed our deep learning algorithms for enhancing healthcare informatics. In Chapter 3, we present a novel “dilated CNN+RN” which combines a dilated CNN with a relation network (RN) to deeply reason the dependencies of non-local axonal connections using whole brain connectome data from children suffering from Focal Epilepsy (FE). In Chapter 4, we present a novel Spatial-Temporal Synchronous Graph Transformer network (STSGT) to capture the complex spatial and temporal dependency of the COVID-19 time series data and forecast the future status of an evolving pandemic. In Chapter 5, we propose a novel dual self-distillation (DSD) framework in U-shaped networks for volumetric medical image segmentation which is a generalized training strategy that utilizes distillation across the encoder and decoder layers of U-Nets extensively to further improve its segmentation performance. In Chapter 6, we address the problem of Seizure Onset Zone (SOZ) localization for children with epilepsy with missing MRI sequence data, by proposing a novel Sequence-Agnostic (SA) model with cross-sequence distillation across the MRI sequence outputs both at the class level and at the feature map level to improve the representation learning of each individual MRI sequence. Finally, in Chapter 7, we conclude the dissertation by highlighting our original contributions for enhancing healthcare informatics with deep learning and discuss some promising future work in this direction.
Recommended Citation
Banerjee, Soumyanil, "Enhancing Healthcare Informatics Through Deep Learning With Graph-Based Models And Self-Distillation" (2024). Wayne State University Dissertations. 4033.
https://digitalcommons.wayne.edu/oa_dissertations/4033
