Application of Different Strategies in Drawing Classification for Parkinson's Disease Diagnosis

The expanding body of research in this field underscores the growing interest in Spiral Analysis (SA) as an effective tool for the detection and monitoring of Parkinson’s Disease (PD). SA-based methods present a cost-effective and accurate alternative to conventional diagnostic techniques, including clinical assessments, laboratory tests, and neuroimaging, which are often time-consuming, expensive, and less accessible. Due to its simplicity and affordability, SA is particularly well-suited for broader clinical implementation, especially in resource-limited settings. This study leverages these advantages by analyzing spiral drawings to extract specific features—such as tremor-induced distortions and area measurements—through an algorithm that evaluates the coordinates of each drawing. These features were used to train and evaluate four pre-trained Convolutional Neural Networks (CNNs): VGG16, ResNet50, MobileNetV2, and Xception. The models were developed and optimized using MATLAB’s Deep Learning Toolbox to classify participants as either belonging to the Parkinson’s Disease (PD) group or the Control Group (CG), and to identify different stages of disease progression within the PD group based on the extracted features. The analysis demonstrates that feature-based classification can effectively support both PD detection and staging. Moreover, the rapid advancement of computer vision techniques further enhances the potential of this approach as a scalable and practical solution for diagnosis, monitoring, and rehabilitation in PD.