Cervical Dystonia Detection using Facial and Eye Feature

This paper proposes a deep learning and traditional machine learning based automatic fusion detection method for Spasmodic Torticollis (the most common type of Cervical dystonia), a neurological disorder. The proposed method utilizes videos of subjects where all of the subjects will be tested if they have Cervical Dystonia or not. For Neurological disorders, generally, very less data is available in public domain due to patient anonymity issue. The paper focused on training Cervical dystonia detection model on very less dataset. Deep learning in the methodology is used to detect the features providing information to traditional ML models for classification task. Methodology developed can be also be extended to grade the severity of disorder. The proposed model achieves video classification accuracy of 90.00% using SVM as final traditional machine learning classifier. We also contribute the first publicly available dataset for Cervical dystonia.