Visual Computing and AI Technologies to Analyse Movement Disorders Associated with Parkinson’s Disease for Diagnostic Purposes

Clinical assessments of Parkinson’s disease (PD) centre on clinicians’ inherently subjective visual interpretations of characteristic motor signs like bradykinesia. This research investigates comprehensive video and signal processing techniques and their capacity to yield clinically meaningful representations of PD movement disorders. PD and healthy control participants were video recorded performing three upper limb motor examinations, including finger tapping, opening-closing and pronation-supination hand movements. Clinicians reviewed movement performance in each video and assigned corresponding MDS-UPDRS scores, providing motor symptom severity class labels for classification and correlation analysis. A real-time computer vision method used a custom-trained YOLO model to evaluate finger tapping videos, analysing computer features and their association with clinical ratings using Spearman coefficients. An automated framework employing MediaPipe Hands evaluated all three motor tasks, transforming extracted motion features into principal components for classification. Results indicate promising classification accuracy, effectively discriminating between motor symptom severity levels in both binary (mild/ severe) and multiclass (mild/ moderate/ severe) classifications.

URI

https://bradscholars.brad.ac.uk/handle/10454/20963

Type

Thesis

Qualification name

PhD