Study Rationale:
Clinical research as well as wearable devices and phone apps used by people with Parkinson's disease (PD) generate large volumes of data, such as genetic and imaging data and data on motor symptoms. The extreme complexity and volume of this data call for new, efficient computer analysis methods.
Hypothesis:
Neural-network-based learning is a programming approach inspired by the remarkable ability of the brain to analyse information. This approach has already found application in image and speech recognition. Our goal is to determine whether neural-network-based learning is a better way to analyse extremely large volumes of diverse biomedical data than state-of-the-art biomedical temporal analysis, an alternative data analysis method.
Study Design:
We will analyse movement and clinical data using three neural-network-based methods: reservoir computing, long short-term memory and a delay-line multilayer perceptron. The approach we are planning to use is one of the most advanced ever used in Parkinson's disease research.
Impact on Diagnosis/Treatment of Parkinson's disease:
These analyses will allow us to predict disease progression, track movement changes associated with PD and obtain immediate feedback on the therapeutic efficiency. This, in turn, will make possible close monitoring of disease progression and rapid adjustment of treatment.
Next Steps for Development:
In the near future, we plan to develop ways to predict upcoming disruptive events, such as freezing of gait, and warn people with PD of them. We also will further develop the data analysis tools used within the scope of this project.