Parkinson's Disease EMG Data Augmentation and Simulation with DCGANs and Style Transfer

Area of Science:

• Biomedical Engineering
• Signal Processing
• Machine Learning

Background:

• Parkinson's Disease (PD) diagnosis and treatment monitoring often rely on electromyography (EMG) signals.
• Acquiring diverse and extensive EMG datasets for research and validation is challenging.
• Existing data augmentation techniques may not adequately capture the complexity of PD tremors.

Purpose of the Study:

• To develop advanced data augmentation techniques for Parkinson's Disease (PD) electromyography (EMG) signals.
• To improve the simulation of patient-specific tremor patterns.
• To create a more robust dataset for validating therapeutic interventions.

Main Methods:

• Implementation of Deep Convolutional Generative Adversarial Networks (DCGANs) for EMG signal generation.
• Application of Style Transfer techniques to augment existing EMG data.
• Validation of generated signals against natural tremor characteristics (frequency, amplitude).

Main Results:

• The proposed DCGAN and Style Transfer models effectively augmented PD EMG datasets.
• The models demonstrated adaptability to varying tremor frequencies and amplitudes.
• Simulated tremor patterns accurately reflected individual patient characteristics.

Conclusions:

• The novel data augmentation approaches significantly enhance the available PD EMG datasets.
• These methods provide a powerful tool for generating realistic tremor simulations.
• The generated simulations can be utilized for validating Parkinson's treatment strategies across diverse movement scenarios.