Web-based Visualization and Machine Learning in Science and Medicine

PhD Thesis Defense by Loraine Franke

Committee Members: Prof. Daniel Haehn (Chair), Prof. Funda Durupinar Babur, Prof. Ping Chen, Prof. Dan Simovici

This thesis explores integrating advanced web technologies with machine learning for scientific visualization. We demonstrate innovative applications of WebGL, WebXR, and machine learning algorithms to create versatile and accessible visualization tools across various domains. The first study presents an analysis of state-of-the-art web-based visualizations, highlighting the transformative impact of new technologies in rendering complex three- dimensional graphics and immersive augmented/virtual reality environments, and introduces the Scientific Visualization Future Readiness Score (SciVis FRS) to assess their preparedness for technological evolution.

Secondly, we propose FiberStars, a visual analysis tool designed for comparative study of diffusion MRI data. The tool uniquely combines three-dimensional anatomical visualization with compact two-dimensional representations, enabling a more effective comparison of brain connectivity patterns across different subjects and groups. We perform a user study with experts and non-experts to test usability and performance of our system.

In the third project, we introduce SlicerTMS, a real-time visualization system for treatment planning in Transcranial Magnetic Stimulation (TMS), a non-invasive neuromodulation technique for treating various brain disorders and mental health diseases. By integrating deep learning, this tool revolutionizes TMS treatment by enabling rapid, precise predictions of electric field distributions, thereby enhancing the effectiveness of brain stimulation therapies. We evaluate our tool with three studies: a real-world use case scenario in a clinical setting, a comparative analysis with another TMS tool focusing on computational efficiency and an expert user study to measure our system’s usability.

Lastly, the fourth work introduces AutoRL X, an open-source, web-based platform for automated reinforcement learning. Building upon the foundation of AutoDOViz, AutoRL X offers an intelligent interface with real-time visualization capabilities, enhancing the understanding and application of reinforcement learning in complex domains like healthcare. Collectively, these studies highlight the importance of visualization and machine learning for complex real-world problems, contributing to advances in scientific research, medical treatment planning, and innovative solutions in various fields.