My paper was submitted! See the publications tab for more about this paper.
No specific test exists to diagnose Parkinson's disease and there is no cure. My dissertation research directly addresses the critical need for early non-invasive diagnostics and biomarkers for the detection and treatment of Parkinson’s disease (PD). This work was conducted utilizing the largest and most robust PD dataset available (a). The PPMI dataset is a landmark longitudinal study that has collected data from more than 1,400 individuals at 33 sites in 11 countries. Parkinson’s disease research is currently data-rich but insight light. To exploit this new wealth of data I generated a well-curated, clinically relevant, relational database constructed specifically for building machine learning training and testing sets.
Machine learning classifiers trained on this dataset were able to predict PD with high accuracy and precision. This analysis found a clear genetic signature that was not significantly impacted by the inclusion of participant age. Parkinson’s disease is believed to be multifactorial, consisting of genetic, environmental, and age-related factors. This analysis showed the need for feature selection methods when applying deep learning methods to these multifactorial data. This study also found that a significant number of misclassified samples are those who switched classes from healthy control to PD mid-study. These patients also had no neurological symptoms upon primary diagnosis. These misclassified samples could hold insights into the progression of PD and early detection of the disease. The conclusion of my work identified bottlenecks in model development, the need for standardized datasets, and a way to compare model performance.
In addition to the contributions described above, with a team of collaborators and community partners, I directly documented the needs of mobility-impaired students on the campus of George Mason University and designed a minimal viable product to address these needs. The result of this year-long community-engaged interdisciplinary design project was a wayfinding app designed to aid mobility-impaired visitors to find acceptable routes around campus. This project included a data-collection robot, a server-side database, and an application. This work demonstrates the need for standardized methods of collecting and mapping GIS data for footpaths to meet the needs of ADA regulations and community members.