A Deep Learning Framework for Early Parkinson's Disease Detection: Leveraging Spiral and Wave Handwriting Tasks with...

Area of Science:

• Neurology
• Artificial Intelligence
• Biomedical Engineering

Background:

• Early detection of Parkinson's disease (PD) is crucial for effective management and improved patient outcomes.
• Current diagnostic methods for PD often rely on subjective clinical assessments, highlighting the need for objective and accessible tools.
• Handwriting impairments are recognized as an early symptom of PD, making handwriting analysis a potential avenue for early detection.

Purpose of the Study:

• To develop and evaluate a novel deep learning framework for the early detection of Parkinson's disease (PD).
• To assess the efficacy of analyzing spiral and wave handwriting patterns for PD diagnosis using advanced AI techniques.
• To establish a robust and generalizable method for objective PD detection through handwriting analysis.

Main Methods:

• A deep learning framework was developed utilizing the PaHaW dataset, focusing on spiral and wave handwriting patterns.
• A preprocessing pipeline involving histogram equalization and Canny edge detection was applied to the handwriting images.
• The framework employed a fine-tuned EfficientNetV2-S architecture for binary classification, alongside a baseline Convolutional Neural Network (CNN) model, with evaluation via 5-fold cross-validation.

Main Results:

• The deep learning models demonstrated high validation accuracies, achieving 98.68% for spiral patterns and 98.10% for wave patterns.
• Excellent Receiver Operating Characteristic-Area Under the Curve (ROC-AUC) scores were obtained, indicating strong discriminatory power between healthy individuals and those with PD.
• Consistent sensitivity and specificity were observed, with a low standard deviation of ±0.0109 from 5-fold cross-validation, confirming model robustness.

Conclusions:

• Handwriting analysis, particularly of spiral and wave patterns, shows significant potential as an objective tool for the early detection of Parkinson's disease.
• The proposed deep learning framework offers a promising, accurate, and reliable method for non-invasive PD diagnosis.
• This approach could complement existing diagnostic methods, enabling earlier intervention and potentially improving patient prognosis.