A Template-Independent Method for Parkinson’s Diagnosis from Handwritten Patterns

Parkinson’s Disease (PD) is a chronic and progressive neurodegenerative disorder that primarily impairs motor function. In recent years, advancements in deep learning (DL) have opened promising avenues for early, non-invasive diagnosis of PD through the analysis of handwriting patterns. This study investigates the use of Convolutional Neural Network (CNN) and Vision Transformers (ViT) models for classifying hand-drawn spiral and meander images collected from both healthy individuals and PD patients. We introduce a robust preprocessing pipeline tailored to enhance diagnostic features and implement a patient-exclusive data partitioning strategy to ensure clinical validity, an aspect often overlooked in prior studies. Our experiments evaluate multiple DL architectures under different fine-tuning strategies, and results show that the proposed preprocessing leads to noticeable improvements in classification accuracy. Despite more rigorous and realistic evaluation protocols, our models achieve performance levels comparable to or exceeding those in existing literature, highlighting both the effectiveness and clinical applicability of our approach.