Vocal Feature Extraction-Based Artificial Intelligent Model for Parkinson's Disease Detection.

Area of Science:

• Neurology
• Computational Intelligence
• Biomedical Engineering

Background:

• Parkinson's disease (PD) is a neurodegenerative disorder impacting nerve cells.
• Early detection of PD is crucial for symptom management.
• Vocal impairments are recognized as early indicators of PD.

Purpose of the Study:

• To propose and evaluate two hybrid models for Parkinson's disease detection using vocal features.
• To compare the efficacy of a Sparse Autoencoder-Support Vector Machine (SAE-SVM) model against a Principal Component Analysis-Support Vector Machine (PCA-SVM) model.
• To assess the performance of these models on imbalanced datasets.

Main Methods:

• Developed two hybrid models: PCA-SVM and SAE-SVM.
• Utilized vocal features for PD detection.
• Employed Support Vector Machine (SVM) for classification.
• Implemented Sparse Autoencoder (SAE) with L1 regularization for feature compression.
• Applied Principal Component Analysis (PCA) for feature reduction.
• Used SMOTE for oversampling and balancing the dataset.

Main Results:

• The SAE-SVM model achieved the highest accuracy (0.935), F1-score (0.951), and Mathews Correlation Coefficient (MCC) (0.788).
• The SAE-SVM model outperformed the PCA-SVM model, along with other standard models (MLP, XGBoost, KNN, RF).
• The proposed models surpassed results from two recent studies using the same dataset.
• Dataset balancing with SMOTE significantly improved model performance.

Conclusions:

• The SAE-SVM hybrid model is highly effective for early Parkinson's disease detection via vocal analysis.
• Deep learning approaches like SAE offer significant advantages in feature extraction for PD detection.
• Vocal biomarkers combined with advanced machine learning techniques provide a promising avenue for non-invasive PD diagnosis.