Comparative Motor Pre-clinical Assessment in Parkinson’s Disease Using Supervised Machine Learning Approaches

• Published in: Annals of biomedical engineering Vol. 46; no. 12; pp. 2057 - 2068
• Main Authors: Rovini, Erika, Maremmani, Carlo, Moschetti, Alessandra, Esposito, Dario, Cavallo, Filippo
• Format: Journal Article
• Language: English
• Published: New York Springer US 01.12.2018; Springer Nature B.V
• Subjects: Adults; Aged; Artificial intelligence; Biochemistry; Biological and Medical Physics; Biomedical and Life Sciences; Biomedical Engineering and Bioengineering; Biomedicine; Biophysics; Classical Mechanics; Classification; Feature extraction; Female; Gait - physiology; Humans; Inertial sensing devices; Learning algorithms; Machine learning; Male; Middle Aged; Motors; Movement disorders; Neurodegenerative diseases; Olfaction disorders; Olfaction Disorders - diagnosis; Olfaction Disorders - physiopathology; Parameter identification; Parkinson Disease - diagnosis; Parkinson Disease - physiopathology; Parkinson's disease; Patients; Quality of life; Supervised Machine Learning; Decision support systems; Idiopathic hyposmia; Supervised learning; Motion analysis; Inertial wearable sensors
• ISSN: 0090-6964; 1573-9686; 1573-9686
• DOI: 10.1007/s10439-018-2104-9

Millions of people worldwide are affected by Parkinson’s disease (PD), which significantly worsens their quality of life. Currently, the diagnosis is based on assessment of motor symptoms, but interest toward non-motor symptoms is increasing, as well. Among them, idiopathic hyposmia (IH) is associated with an increased risk of developing PD in healthy adults. In this work, a wearable inertial device, named SensFoot V2, was used to acquire motor data from 30 healthy subjects, 30 people with IH, and 30 PD patients while performing tasks from the MDS-UPDRS III for lower limb assessment. The most significant and non-correlated extracted parameters were selected in a feature array that can identify differences between the three groups of people. A comparative classification analysis was performed by applying three supervised machine learning algorithms. The system resulted able to distinguish between healthy and patients (specificity and recall equal to 0.967), and the people with IH can be identified as a separate class within a three-group classification (accuracy equal to 0.78). Thus, the system could support the clinician in objective assessment of PD. Further, identification of IH together with changes in motor parameters could be a non-invasive two-step approach to investigate the early onset of PD.