Thesis labs: Rajat Dar and Daniel Marcus (WashU Radiology)
Abstract: One of the main restrictions in medicine today is our limited biological understanding of complex diseases such as stroke and obesity. Further knowledge discovery can provide effective biomarkers to improve disease diagnosis and prognosis, identify genetic drivers, predict individual genetic susceptibility for effective disease management, and develop personalized drugs. Big data contributes to knowledge discovery through the development of centralized data management systems with the capability to enable multi center collaborators to jointly share, manage, and collect massive multimodal data (imaging, multi-omics, and clinical data) from hospitals; develop and launch containerized Artificial Intelligence (AI) processing pipelines at scale for knowledge discovery; and validate AI-based pipelines on standard heterogeneous datasets before deployment in clinical settings as medical device. In this thesis, I describe three key directions that present opportunities for the development of informatics and deep learning technologies in medicine. Initially, I discuss the development of the Stroke Neuro-Imaging Phenotype Repository (SNIPR), an XNAT-based open data science platform, for multi-center stroke research. The development of AI applications for curation of CT and MRI brain scan types is also explained to automate scan selection for a broader pipeline development strategy in SNIPR. Secondly, I detail a large multi-center longitudinal study, with a focus on predicting malignant cerebral edema using clinical phenotypes and imaging phenotypes derived from SNIPR data. Lastly, I discuss the application of AI to predict polygenic phenotypes and measure pointwise uncertainty from genomics data.