Framework of GCN for Parkinson's Detection

Parkinson’s Disease (PD) is a neurodegenerative disorder that progressively impairs motor and speech functions, reducing quality of life. Early and accurate detection can help manage its symptoms more effectively through remote clinical monitoring. This study introduces a lightweight Graph Convolutional Network (GCN) architecture designed to detect PD from small voice datasets, emphasizing relation over repetition and structure over scale. Voice samples are represented as nodes within a similarity-driven graph, allowing information to propagate across related instances rather than being treated as isolated points. Through stratified 5-fold validation, the model achieved an accuracy of 0.7772, outperforming recent baselines while maintaining interpretability. The conceptual design originated from an analogy-driven insight—comparing voice coordination to an orchestra where harmony arises from relation, not volume—while technical implementations and validation were supported by AI reasoning systems under human supervision. The results demonstrate that structured connectivity can substitute for model depth, promoting a new paradigm of transparent, collaborative, and data-efficient clinical AI.