Enhancing parkinson disease detection through feature based deep learning with autoencoders and neural networks

Area of Science:

• Neuroscience
• Artificial Intelligence
• Biomedical Engineering

Background:

• Parkinson's disease (PD) is a progressive neurodegenerative disorder linked to aging, impacting brain regions and necessitating accurate diagnosis for effective therapy.
• Current diagnostic methods may lack the precision and timeliness required for optimal patient outcomes.
• Early and non-intrusive detection of PD is crucial for improving treatment efficacy and patient quality of life.

Purpose of the Study:

• To develop and evaluate an innovative approach for the automated and non-intrusive identification of Parkinson's disease (PD) using audio wave analysis.
• To leverage Feature-Based Deep Neural Network (FB-DNN) techniques, integrating Autoencoder for feature extraction and Deep Neural Networks (DNNs) for classification.
• To enhance diagnostic accuracy and enable prompt identification of PD through subtle voice characteristic variations.

Main Methods:

• Utilized Autoencoder, a type of Artificial Neural Network (ANN), for effective feature extraction from audio data.
• Employed Deep Neural Networks (DNNs) for the classification task, differentiating between audio samples indicative of PD and healthy controls.
• Trained the DNN model on audio data to recognize subtle voice variations associated with Parkinson's disease.

Main Results:

• The Feature-Based Deep Neural Network (FB-DNN) approach demonstrated superior performance compared to other models.
• The FB-DNN model achieved a high accuracy score of 96.15% in identifying Parkinson's disease.
• The study successfully implemented the methodology in Python, validating its practical application.

Conclusions:

• The integration of Autoencoder-based feature extraction with DNNs offers a dependable and accessible solution for early PD detection and monitoring.
• This audio analysis methodology shows promise for significantly improving the quality of life for individuals with Parkinson's disease.
• The FB-DNN technique presents a potentially effective strategy for automated, non-intrusive Parkinson's disease diagnosis.