Parkinson’s Disease Detection Through Static Handwriting Analysis Using CNNs and SVM Ensemble

Parkinson's Disease is a progressive neurological disorder marked by motor impairments, often reflected in handwriting anomalies such as tremors, micrographia, and stroke irregularities. The diagnosis often relies on subjective clinical evaluation. This study explores an objective detection framework using static handwriting samples, specifically spiral and wave drawings. The method employs an ensemble strategy that fuses deep features from two pre-trained Convolutional Neural Network (CNN) architectures, ResNet50 and DenseNet121, and classifies the feature representations using a Linear Support Vector Machine (SVM). To ensure rigor, a strict split-before-augmentation protocol was applied to a primary dataset of 204 subjects. The model achieved 97.28% accuracy on the primary test set. When independently verified using a local dataset, the system maintained strong generalization with an accuracy of 95.19%, 97.04% PD recall, and 93.33% specificity. Statistical analysis confirmed these results with narrow confidence intervals, validating the system's stability. The results confirm that dual-CNN feature fusion significantly improves detection stability and adaptability between datasets. This framework serves as an objective, scalable screening tool to complement neurological assessments.