Objective: Primary Objective:

To develop and evaluate a machine learning-based model for the accurate classification of tremor and dystonia using video data.

Secondary Objectives:

To address the shortage of movement disorder specialists in hospitals across Egyptian governorates and African countries.

Background: Movement disorders like dystonia and tremors significantly impact patients’ quality of life and healthcare systems. Dystonia, the third most common movement disorder, while essential tremor affects up to 5.6% of the global population. Delayed diagnosis, often taking 3-8 years, leads to worsened symptoms and psychological comorbidities. This challenge is further compounded by a global shortage of neurologists, especially in Africa and Asia, limiting access to timely treatment and care. This study leverages deep learning models integrated into a mobile application to enhance the classification and early detection of movement disorders.

Method: A real dataset of 100 patient videos (50 tremor and 50 dystonia cases) was used from the records of the Movement Disorders Clinic at Ain Shams University. Videos were pre-processed with facial blurring to ensure patient confidentiality. The models used included CNN-LSTM, CNN-RNN, 3D CNN, and ResNet-RNN. Data augmentation techniques were applied, and the dataset was split into training and testing sets. Accuracy, sensitivity, and specificity were evaluated.

Results: The ResNet-RNN model achieved the highest accuracy (85.24%), followed by CNN-LSTM (82.46%). The deep learning models effectively distinguished between tremor and dystonia, demonstrating the potential to support early diagnosis and reduce diagnostic delays.

Conclusion: AI-driven applications in movement disorders can aid in screening and serve underserved areas. The proposed deep learning models show significant potential in revolutionizing the diagnosis, treatment, and management of tremor and dystonia. Ongoing research and development are essential for improving these AI models and integrating them into clinical practice. Future work will focus on expanding the dataset and incorporating other movement disorders for improved generalization.

« Back to 2025 International Congress

MDS Abstracts - https://www.mdsabstracts.org/abstract/ai-based-detection-of-tremor-and-dystonia-using-deep-learning-models-a-novel-approach/