Objective: To create a quantitative, objective framework for assessing combined dystonia parkinsonism using kinematic data from wearable sensors combined with machine learning (ML), enabling clinician-independent identification and severity evaluation.

Background: X-linked dystonia parkinsonism (XDP) is a rare neurogenetic combined movement disorder, with prominent dystonia and parkinsonian features, resulting in complex, overlapping motor symptoms that challenge traditional assessment. Objective, rater-independent motor assessments are crucial for improving diagnosis and monitoring change for XDP and other forms of combined parkinsonism.

Method: 28 patients with manifest XDP and 5 healthy controls were assessed using 17 wearable inertial measurement unit sensors and motion sensor gloves with a standardized motor examination. Dystonia and parkinsonian features were assessed with the Unified Parkinson’s Disease Rating Scale (MDS-UPDRS) and Burke-Fahn-Marsden scale (BFM). We crafted feature extraction from sensor data to mirror clinicians’ observations during assessments and applied the Synthetic Minority Over-sampling Technique to address class imbalance. Sammon mapping projection algorithms were applied to explore the feature space, and random forest algorithms were used to identify and quantify dystonic and parkinsonian features and predict clinical scores.

Results: Across all tasks analyzed, including those involving the upper limbs (e.g., finger tapping), lower limbs (e.g., toe-tapping), and gait, the features obtained from sensor data demonstrated clear cluster differentiation between control participants and individuals with XDP. The use of a random forest regressor achieved an accuracy of up to 85% in differentiating between the two groups. Furthermore, projections identified distinct clusters of sensor-derived features corresponding to various MDS-UPDRS scores in the tasks assessed. The prediction of MDS-UPDRS scores using random forest regression algorithms displays an accuracy significantly and consistently above the chance level across tasks, indicating that the features extracted from sensor data provide a reliable basis for estimating clinical scores.

Conclusion: This approach underscores the utility of wearable technology and ML in characterizing dystonic and parkinsonian motor features, paving the way for clinician-independent assessment in XDP and other causes of combined dystonia.

« Back to 2024 International Congress

MDS Abstracts - https://www.mdsabstracts.org/abstract/quantitative-assessment-of-x-linked-dystonia-parkinsonism-using-wearable-sensing-technology/