Graph-Based Deep Feature Extraction for Automated Parkinson's Disease Recognition

Abstract

Parkinson's Disease (PD) is a progressive neurodegenerative disorder that affects millions of individuals worldwide and is characterized by motor and non-motor impairments, including tremors, rigidity, bradykinesia, postural instability, cognitive decline, and speech abnormalities. Early and accurate diagnosis of Parkinson's Disease is essential for effective treatment planning, disease management, and improvement of patient quality of life. However, conventional diagnostic approaches largely depend on clinical observation and neurological assessments, which may be subjective and prone to variability among healthcare professionals. Consequently, there is an increasing demand for intelligent automated diagnostic systems capable of identifying Parkinsonian patterns with high accuracy and reliability. This study proposes a Graph-Based Deep Feature Extraction Framework for Automated Parkinson's Disease Recognition (GDFE-PDR) that integrates graph modeling techniques and advanced deep learning architectures for intelligent disease classification. The proposed framework transforms biomedical and neurological data into graph representations to capture complex structural relationships among disease-related features. Graph-based feature extraction mechanisms effectively preserve connectivity patterns and hidden dependencies, while deep learning models learn discriminative representations associated with Parkinsonian abnormalities. Furthermore, adaptive graph learning enhances robustness against noisy and heterogeneous clinical data. Experimental evaluation demonstrates substantial improvements in classification accuracy, sensitivity, specificity, precision, and F1 score when compared with traditional machine learning and conventional deep learning approaches.