Study Rationale:
In-home monitoring of the severity of Parkinson's disease (PD) may help guide therapeutic decisions and prevent life-threatening falls. Deep learning is the state-of-the-art tool for analyzing time-series data. This technique teaches computers to classify items. The aim of this study is to develop deep learning models for the accelerometer data collected through the Michael J. Fox Foundation-supported Clinician Input Study, which aimed to better understand the use of wearable devices in clinical care, and correlate the predictions to the clinical outcomes. Accelerometers measure acceleration or movement.
Hypothesis:
We hypothesize that deep learning can extract important signals from gait records collected from wearable devices, which can be further correlated with clinical outcomes.
Study Design:
Firstly, we will transfer machine learning models built on the mobile Parkinson's disease study mPower to the Clinician Input Study data. Both studies have accelerometer data, which makes these two data sets comparable to each other. mPower data are much diversely sampled -- from more than 3,000 unique participants -- so we expect the models to be robust and generalizable to unseen individuals. We will correlate the motion-based features in deep learning networks with clinical features to obtain an understanding of the models and Parkinson's monitoring. We will also train models using only the Clinician Input Study data. Although the number of unique samples is relatively small, we can segment the long stretches of 24/7 record into short chunks, generating large amount of training examples.
Impact on Diagnosis/Treatment of Parkinson's Disease:
The study will provide us a deeper understanding of the Clinician Input Study data. These findings could lead to greater understanding of how wearable devices can help measure Parkinson's. And the developed deep learning models could be applied to other studies.
Next Steps for Development:
The direct outcome of the study is to help understand data we have already collected and help design future wearable device studies.