Detecting Freezing-of-Gait Symptom in Parkinson’s Disease by Analyzing Vertical Motion from Force Plate

• Published in: New Frontiers in Artificial Intelligence pp. 468 - 477
• Main Authors: Le, Dinh-Khiet, Torii, Takuma, Fujinami, Tsutomu, Buated, Wannipat, Lolekha, Praween
• Format: Book Chapter
• Language: English
• Published: Cham Springer International Publishing
• Series: Lecture Notes in Computer Science
• Subjects: Cognitive loading; Freezing of gait; Parkinson’s disease; Postural control
• ISBN: 9783030316044; 3030316041
• ISSN: 0302-9743; 1611-3349
• DOI: 10.1007/978-3-030-31605-1_34

Introduction: Freezing of Gait (FoG) is a common symptom in Parkinson’s Disease (PD), which has impact on the gait pattern and relevant to risk of falls. Data-driven approach to FoG detection would allow systematic assessment of patient’s condition and objective evaluation of the clinical effects on treatments. Many researchers recently studied FoG in PD by analyzing patient’s center of pressure dynamics in term of various features such as path-length. Objective: In this research, we attempt to automatically classify two groups of PD patients that with and without FoG by considering standing balance ability during cognitive loading tasks. Methods: The dataset consists of sixty PD patients (Hoehn and Yahr stages 1–3) were collected from Thammasat University Hospital, Thailand. The participants were categorized either to be FoG or non-FoG according to the Freezing of Gait-Questionnaire (FoG-Q) scores. Their postural balance ability was measured with Nintendo Balance board which produces a time-series of center of pressure along with the value of changing weight. We turn to a new kind of feature named “Fluctuation of Vertical Acceleration” (FVA) which informs us the acceleration due to the body’s up-down motion and use comparative analysis to analyze the postural control function activities in cognitive loading tasks of all patients, FoG and non FoG groups. Results: Significant increases of the FVA were observed when applying cognitive loading (p < 0.001) in all cases (considering all data or each subgroup). The FVA also increased between the rest state and the other rest state after a cognitive loading task (p < 0.001). The difference between FoG and non FoG was observed by using FVA (p < 0.05). The test results when using FVA are in line with using other features extracted from the trajectory of center of pressure (such as path-length). Conclusions: The new simple feature, FVA, seems to reflect well postural control activities in people with PD, especially recognizing the change in a cognitive loading task. In addition, based on the postural control function, indirectly through the FVA, we are possible to classify automatically PD patients into the FoG or the non-FoG group.