Simulating Accelerometer Signals of Parkinson's Gait Using Generative Adversarial Networks

Area of Science:

• Biomedical Engineering
• Neurology
• Machine Learning

Background:

• Wearable sensors offer objective Parkinson's disease (PD) assessment.
• Monitoring freezing of gait (FOG) in PD patients remains challenging with current methods.
• Data augmentation enhances machine learning model accuracy in healthcare.

Purpose of the Study:

• To evaluate generative adversarial networks (GANs) for creating synthetic accelerometry data.
• To generate realistic gait patterns, including typical and FOG, for individuals with PD.
• To explore a novel approach for improving FOG detection algorithms.

Main Methods:

• Utilized generative adversarial networks (GANs) to synthesize accelerometry data.
• Focused on generating data representing both normal and freezing of gait (FOG) patterns.
• Simulated movement data for individuals with Parkinson's disease.

Main Results:

• Preliminary results indicate synthetic datasets mimic realistic PD movement patterns.
• Generated data shows potential for capturing nuances of typical and FOG gait.
• The synthetic data successfully replicated key characteristics of patient movement.

Conclusions:

• Generative adversarial networks show promise for creating synthetic gait data in Parkinson's disease.
• This synthetic data may enhance the training of machine learning models for FOG detection.
• Further research will validate the impact of synthetic data on FOG detection algorithm accuracy.