DEVELOPMENT OF A NON-INVASIVE AND COST-EFFECTIVE SCREENING TOOL FOR THE DETECTION OF PARKINSON’S DISEASE: A PRELIMINARY STUDY

Abstract

Parkinson's disease (PD) is a progressive neurodegenerative disorder affecting movement that typically remains undiagnosed until noticeable clinical symptoms emerge. Accessible home-based screening tools are essential for early PD detection. This study developed an artificial intelligence classification model to detect Parkinson's disease from spiral drawings as a motor assessment using Teachable Machine. The model was trained on a public dataset of 1,590 images, divided into training (765 images per class, N = 1,530) and testing sets (30 images per class, N = 60). The Teachable Machine model achieved high classification performance with 98.33% accuracy, 100% precision, and 96.77% recall. External validation using 20 images from healthy subjects (aged 17-76 years) yielded 80% accuracy. These results demonstrate the model's effectiveness in distinguishing between Parkinson's disease and healthy hand-drawn spiral images. Despite evidence of overfitting, the model shows potential as a preliminary screening tool for PD detection. Further validation studies with larger, more diverse datasets are warranted.