FoGGAN: Generating Realistic Parkinson's Disease Freezing of Gait Data Using GANs

Area of Science:

• Artificial Intelligence in Healthcare
• Medical Data Augmentation
• Neurological Disorder Research

Background:

• Healthcare data scarcity hinders AI development due to collection and privacy challenges.
• Parkinson's Disease (PD) data acquisition is difficult, impacting AI model training.
• Data augmentation is crucial for advancing AI in medicine.

Purpose of the Study:

• To develop a Generative Adversarial Network (GAN) based data augmentation solution for Parkinson's Disease (PD) freezing of gait (FoG) data.
• To evaluate the effectiveness of synthetic data generated by the proposed FoGGAN architecture.
• To assess the performance of a Deep Neural Network (DNN) classifier trained on augmented data.

Main Methods:

• Implementation of a novel Generative Adversarial Network (GAN) architecture, termed FoGGAN.
• Utilizing a freezing of gait (FoG) symptom dataset as input for the FoGGAN model.
• Employing various similarity metrics to compare generated data with original data.
• Evaluating a Deep Neural Network (DNN) classifier on three distinct datasets.

Main Results:

• The FoGGAN architecture successfully generated synthetic data highly similar to the original FoG dataset, confirmed by similarity metrics.
• The generated data proved credible for use in AI training datasets.
• A DNN classifier demonstrated encouraging accuracy when evaluated on datasets including the synthetically generated data.

Conclusions:

• The FoGGAN solution effectively addresses the data shortage issue in PD research.
• GAN-based data augmentation provides a viable method for creating high-quality synthetic medical data.
• This approach can significantly support the development and deployment of AI applications in healthcare, particularly for neurological conditions.