Deep Learning-Based Diagnostic Model for Parkinson's Disease Using Handwritten Spiral and Wave Images

Area of Science:

• Neurology
• Computer Science
• Medical Imaging

Background:

• Parkinson's Disease (PD) diagnosis can be challenging, requiring objective biomarkers.
• Handwritten spiral and wave drawings offer potential for non-invasive PD assessment.

Purpose of the Study:

• To develop and validate a deep neural network (DNN) model for diagnosing Parkinson's Disease (PD) using handwritten spiral and wave images.
• To compare the DNN model's diagnostic performance against various machine learning (ML) and deep learning (DL) models.

Main Methods:

• A dataset of 204 spiral and wave images from PD patients and healthy subjects was utilized.
• Images were preprocessed using Histogram of Oriented Gradients (HOG) and augmented for diversity.
• A custom DNN architecture was designed and compared against nine ML and two DL models.

Main Results:

• The DNN model demonstrated superior performance in diagnosing PD from handwritten images compared to all other evaluated models.
• Significant improvements in accuracy, sensitivity, and specificity were observed with the DNN model.
• The DNN model achieved notable accuracy gains over models like Naïve Bayes, Decision Tree, and Support Vector Machine on both spiral and wave images.

Conclusions:

• The developed DNN model offers a highly accurate and promising tool for early Parkinson's Disease detection.
• This image-based diagnostic approach provides a foundation for future research incorporating additional clinical features.
• The study highlights the potential of AI in revolutionizing neurological disorder diagnostics.