Parkinson's disease detection based on features refinement through L1 regularized SVM and deep neural network.

Area of Science:

• Biomedical Engineering
• Computational Neuroscience
• Speech Science

Background:

• Previous Parkinson's disease (PD) detection studies often suffer from low accuracy and unreliable results due to inappropriate validation methods.
• The generalization capability of existing PD detection models is questionable due to the methodologies employed.

Purpose of the Study:

• To critically analyze the impact of validation methodologies on PD detection accuracy.
• To propose and evaluate an enhanced two-stage diagnostic system for improved PD detection.
• To ensure the reliability and generalizability of PD detection results.

Main Methods:

• A two-stage diagnostic system was developed, incorporating feature refinement using regularized linear support vector machines.
• Classification of refined features was performed using a deep neural network.
• The system was rigorously evaluated on two benchmark voice recording datasets using Leave-One-Subject-Out (LOSO) and k-fold cross-validation (CV).

Main Results:

• The proposed system achieved 100% accuracy with LOSO CV and 97.5% accuracy with k-fold CV on both datasets.
• The diagnostic system demonstrated superior performance compared to existing methods for PD detection.
• The findings highlight the effectiveness of the proposed system in enhancing PD detection accuracy.

Conclusions:

• The proposed two-stage diagnostic system offers a significant improvement in Parkinson's disease detection accuracy.
• Appropriate validation methodologies are crucial for ensuring the reliability of PD detection results.
• This system holds promise for advancing non-invasive diagnostic decision support for Parkinson's disease.