"Differentiating Stroke and Healthy Adults Using Visualization Driven Timeseries Extraction Method" data files
Rahul Soangra and Joby John
The first file (which can be downloaded via the "Download Data Set" button) contains the OMX data files for healthy adults (Healthy_Raw_OMX_Files.zip).
The additional files include:
- 4 ZIP folders of h5 data files for healthy adults
- 1 ZIP folder of OMX data files for adults who've had strokes
- 1 ZIP folder of h5 data files for adults who've had strokes
- 1 ZIP folder of additional PY, IPYNB, and PDF files (Codes_DataProcess.zip)
Wearable technologies may offer ways to measure unhindered activities of daily living (ADL) among patients who had stroke in their natural settings. This study investigated new visualization driven timeseries extraction methods for classification of activities from stroke and healthy adults. Stroke and healthy adults wore a sensor at L5/S1 area for three consecutive days and data was collected passively in participant’s naturalistic environment. Data visualization resulted in selected timeseries which were classified with an accuracy of 97.3% using recurrent neural network. We found negative correlation between body mass index BMI and higher acceleration fraction. We also found individuals with stroke produced activity amplitudes lower than healthy counterparts in all activity bands (low, medium and high). Our findings demonstrate that intelligently extracted visualization-driven timeseries and a recurrent neural network can accurately classify movements amongst stroke and healthy groups. This use of visualization based timeseries extraction from naturalistic data is a major strength of this study and lays foundation for passive ADL monitoring in real-world environments. Such timeseries extraction methods along with RNN classifiers have potential to track the daily progress of rehabilitation exercises remotely among stroke survivors.
STRIDE is an initiative based at the University of Southern California to create an inter-institutional, public database containing de-identified demographic and kinematic, kinetic, and spatiotemporal measures assessed via gait analysis in individuals post-stroke, to provide a larger and more heterogeneous research dataset than that typically amassed at a single institution. The data in STRIDE can be used to run pilot analyses and power calculations for research studies, design and validate statistical models to test associations between gait variables, provides data for simulation-based biomechanical studies in stroke, and provides data to assess the reproducibility of research findings, without the added data collection requirements.
Supplementary Material to the Manuscript Titled: Mobile Phone Sensors Can Discern Medication-Related Gait Quality Changes in Parkinson's Patients in a Real-World Setting
Niklas König Ignasiak, Albert Pierce, Vincent Berardi, Wilford K. Eiteman-Pang, and Cyril Rakovski
This file contains the data that was used to classify high and low quality gait patterns in patients with Parkinson's disease. Acceleration and gyroscope data was recorded with a conventional smartphone in a real-world environment. High (i.e. ON medication) and low (i.e. OFF medication) quality labels were given by a human observer according to medication intake times.
Supplementary Material to the Manuscript Titled: Self-Paced Treadmills Do Not Allow for Valid Observation of Linear and Non-Linear Gait Variability Outcomes in Patients with Parkinson’s Disease
Niklas König Ignasiak, Maryam Rohafza, Rahul Soangra, and Jo Armour Smith
The data contains measurements on patients with Parkinson's disease and healthy participants walking outdoors, as well as on a treadmill. Stride time and stride length information has been recorded using an inertial measurement unit system. The data allows comparison of linear and non-linear gait variability outcomes from overground walking and treadmill walking for validation purposes.