Gait Analysis for X-linked Dystonia Parkinsonism: Disease Assessment and Severity Estimation Using Sensing Technology and Machine Learning

X-linked dystonia parkinsonism (XDP) is a rare movement disorder seen primarily in individuals from Panay Island, Philippines, marked by adult-onset dystonia that gradually progresses and often transitions into parkinsonism. The aim of this project is to assess the severity of XDP symptoms across four specific conditions (gait, freezing of gait and postural instability) by extracting relevant features from signals acquired through Inertial Measurement Unit (IMU) wearable sensors. The same approach was used to achieve this goal for each of these tasks. The dataset comprises baseline and follow-up data. The baseline includes 5 controls and 32 XDP subjects. Follow-up data were collected from 13 XDP subjects at 6 months and 7 XDP subjects at 12 months after the baseline. Disease severity was assessed with the Unified Parkinson’s Disease Rating Scale Part 3 (MDSUPDRS). Data collections were performed using 17 9-axis IMUs, and suitable data features were derived to estimate clinical scores using machine learning (ML). A certain number of features were selected using Recursive Feature Elimination (RFE), and a feature projection for each lower limb task was performed to inspect the clustering between patients before classification. An ML supervised algorithm was then developed to predict the scores using the Random Forest algorithm with the Leave-One-Out Cross-Validation method. Gait spatio-temporal parameters and frequency-specific features play a crucial role, as the results demonstrate a high degree of clustering between different stages of the disease. However, there is a margin for improvement that could be achieved with a more balanced dataset. This study assessed the feasibility of using wearable sensors to evaluate gait patterns in XDP patients and derived reliable clinical score estimates. The results suggest that wearable technology, combined with advanced feature selection and machine learning algorithms, can be a powerful tool in monitoring and evaluating the progression of movement disorders such as XDP. Further research with larger and more balanced datasets could enhance the accuracy and reliability of these methods, providing valuable insights for clinical assessments.