A Comprehensive Framework for Parkinson's Disease Detection Using Spiral Drawings and Advanced Machine Learning...

Area of Science:

• Neuroscience
• Biomedical Engineering
• Artificial Intelligence

Background:

• Parkinson's disease (PD) diagnosis relies on clinical symptoms, often appearing after significant neurodegeneration.
• Early and accurate detection of PD is crucial for timely intervention and management.

Purpose of the Study:

• To develop a reliable and scalable framework for Parkinson's disease detection using spiral drawings.
• To integrate advanced machine learning techniques for improved diagnostic accuracy in clinical settings.

Main Methods:

• Spiral drawing data from Parkinson's patients and healthy individuals were analyzed.
• Deep learning models (ResNet50, VGG16, EfficientNetB0) extracted features from drawings.
• Feature selection techniques (PCA, RFE, LASSO, ANOVA) and classifiers (SVM, RF, MLP, XGBoost, CatBoost, voting) were evaluated.

Main Results:

• The framework achieved high classification performance, with models reaching up to 98% accuracy and 98% AUC-ROC.
• Specific configurations, such as ResNet50 with PCA and MLP, demonstrated exceptional diagnostic capabilities.
• Ensemble methods and individual classifiers showed robust performance across various metrics.

Conclusions:

• Combining advanced feature extraction, selection, and classification significantly enhances PD detection accuracy.
• The developed framework shows potential for improving the scalability and clinical utility of PD diagnosis.
• Future research should explore multi-modal data integration and real-time applications.