Novel Dual-Stream Deep Ensemble for Parkinson's Disease Detection Using Multi-modal Features from Small Handwritten Dynamics Dataset

Parkinson's disease is one of the most difficult and complex chronic degenerative brain diseases that affects the nervous system. Its symptoms include tremors, muscular rigidity, postural instability, and significant changes in speech and handwriting. Although the disease still lacks a cure, medical advances have demonstrated that early detection and targeted interventions can significantly improve patients' quality of life and slow the progression of symptoms. Researchers in the medical field have explored various data sources for detecting Parkinson's disease, including imaging techniques, clinical records, and dopamine transporter scanning. However, among these methods, less invasive diagnostic tools, especially those utilising subtle patterns in handwriting samples, have been less explored and constitute a potential frontier in diagnostic technology. Our work aims to fill this crucial research gap by presenting a novel deep learning, ensemble-based RXPDNeuroNet, a deep ensemble architecture that combines ResNet50V2 and XceptionNet, for the diagnosis of Parkinson's disease. This architecture is based on an innovative concept of feature concatenation to determine the presence of Parkinson's disease by analysing handwritten records. The proposed model exhibits excellent discriminative power, attaining a remarkable classification accuracy of 94.12%, and the other performance metrics are also in line, with precision, recall, and F1 scores all exceeding 91%. This not only surpasses recent computational diagnostic benchmarks but also establishes a new gold standard in computer-aided medical diagnosis.