Automated Early Detection of Parkinson's Disease Using Graph Convolution Networks

Parkinson's disease affects millions worldwide, emphasizing the critical need for early identification to facilitate effective management as well as treatment. In recent years, the utilization of machine learning methods to enhance the diagnosis of Parkinson's Disease (PD) has gained significant interest. The study proposes an innovative technique for PD detection by integrating Graph Convolutional Networks (GCNs) with handwriting and voice signatures. Initially, stroke features are extracted from handwriting samples, followed by retrieving acoustic features from speech recordings. These extracted features are then used to construct graphs representing each modality's temporal and spatial interactions. The GCN architecture is specifically tailored to process graph-structured data through convolutional and pooling operations. Additionally, to enhance model efficacy, the research investigates the inclusion of attention mechanisms. To enhance model performance, we also investigate integrating attention mechanisms to boost model performance. The experimental results highlight the non-invasive nature and early diagnosis potential of our approach by demonstrating its ability to reliably detect PD using handwriting along with voice data. This research has the potential to revolutionize the early PD diagnosis, leading to more effective management and treatment. Experimental results highlight the effectiveness of our method in accurately detecting Parkinson's disease using handwriting and voice data, with a focus on its non-invasive nature and potential for early diagnosis. With an accuracy of 99.8%, our approach significantly surpasses existing methods, offering great promise for improving early detection and management of Parkinson's disease.