Parkinson’s Disease (PD) is a common neurodegenerative disorder whose clinical picture is characterized by motor and non-motor symptoms. One of motor symptoms is freezing of gait (FoG) that consists in a few seconds during which patients can't start to walk again. In this paper 41 patients affected by PD, with and without FoG, underwent gait analysis performing three gait tasks: normal gait, a motor dual task and a cognitive task. A statistical analysis was performed on clinical, demographical and on the spatial and temporal parameters in order to find any difference between PD patients with and without FoG; the last one obtained no statistically significant results. Thus, a machine learning analysis was implemented employing tree-based algorithms (decision tree, Random Forests, Gradient Boosted Tree, Ada-Boosting of a decision tree) and using as input the spatial and temporal features of gait. The results were promising since accuracy, specificity and sensitivity overcame 90%, reaching also 100% of sensitivity in some cases. The best algorithms were Gradient Boosted Tree and the Ada-Boosting of a decision tree while Random Forests and decision tree obtained lower results. This study proved that machine learning can help to identify patients affected by mild form of FoG that exposes them to a major risk of developing more severe form of freezing with a consequent increased risk of falling.

Classifying patients affected by Parkinson's disease into freezers or non-freezers through machine learning

Cesarelli G.;
2020-01-01

Abstract

Parkinson’s Disease (PD) is a common neurodegenerative disorder whose clinical picture is characterized by motor and non-motor symptoms. One of motor symptoms is freezing of gait (FoG) that consists in a few seconds during which patients can't start to walk again. In this paper 41 patients affected by PD, with and without FoG, underwent gait analysis performing three gait tasks: normal gait, a motor dual task and a cognitive task. A statistical analysis was performed on clinical, demographical and on the spatial and temporal parameters in order to find any difference between PD patients with and without FoG; the last one obtained no statistically significant results. Thus, a machine learning analysis was implemented employing tree-based algorithms (decision tree, Random Forests, Gradient Boosted Tree, Ada-Boosting of a decision tree) and using as input the spatial and temporal features of gait. The results were promising since accuracy, specificity and sensitivity overcame 90%, reaching also 100% of sensitivity in some cases. The best algorithms were Gradient Boosted Tree and the Ada-Boosting of a decision tree while Random Forests and decision tree obtained lower results. This study proved that machine learning can help to identify patients affected by mild form of FoG that exposes them to a major risk of developing more severe form of freezing with a consequent increased risk of falling.
2020
978-1-7281-5386-5
File in questo prodotto:
Non ci sono file associati a questo prodotto.

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: https://hdl.handle.net/11367/137737
Citazioni
  • ???jsp.display-item.citation.pmc??? ND
  • Scopus 14
  • ???jsp.display-item.citation.isi??? 0
social impact