Abstract
Essential tremor (ET) is the most prevalent movement disorder, with its incidence increasing with age, significantly impacting motor functions and quality of life. Traditional methods for assessing ET severity are often time-consuming, subjective, and require in-person visits to medical facilities. This study introduces a novel deep learning-based approach for remotely assessing ET severity using handwriting images, which improves both efficiency and accessibility. We collected approximately 1000 high-quality Archimedean spiral handwriting images from patients in both medical institutions and home settings, creating a robust and diverse dataset. A transfer learning-based model, ETSD-Net, was developed and trained to evaluate ET severity. The model achieved an accuracy of 88.44%, demonstrating superior performance over baseline models. Our approach offers a cost-effective, scalable, and reliable solution for ET assessment, particularly in remote or resource-limited settings, and provides a valuable contribution to the development of more accessible diagnostic tools for movement disorders.