Data-driven analytics, such as machine learning methods, have been applied to discover comprehensive Parkinson’s disease (PD) subtypes from observational patient data. However, many of these approaches have limitations to their clinical utility. The goal of this research is to develop knowledge distillation and visualization approaches.
There are commonalities within the same PD subtype that can be captured by the patient characteristics and their (causal) relationships.
We will enhance the clinical utility of data-driven PD subtypes by 1) mining causal relationships among clinical, neuroimaging and genetic variables; 2) using visualization techniques so that the PD subtypes can be demonstrated graphically to the end users and the end user can interact with the visualization interface to understand the PD subtypes.
Impact on Diagnosis/Treatment of Parkinson’s Disease:
The research in this project will bridge the gap between clinical knowledge and data-driven PD subtypes, which greatly enhances the clinical utility of data-driven algorithms and the subtypes discovered.
Next Steps for Development:
With the clinical knowledge derived from the tools developed in this project, we expect that data-driven PD subtypes can be understood and validated clinically and be helpful in clinical practice.