Gait analysis to quantitatively classify Ataxia and Parkinson’s disease patients: a pilot study using tree-based Machine Learning algorithms

Abstract

Posture and gait anomalies investigations have demonstrated to improve to date the rehabilitation programs and patients’ outcome for several types of disorders [1,2]; however, the diagnosis and progression of some illnesses, such as Ataxia and Parkinson’s Disease (PD), still rely on “gold standard” (clinical) methodologies, which hardly ever allows to fully distinguish the two pathologies. Consequently, the objective of this paper is to provide a new strategy to classify Ataxia and PD subjects using quantitative gait analysis tools. For this purpose, two cohorts of 43 PD and 22 ataxic patients were enrolled. Each patient underwent an Instrumented Stand and Walking [3] test using a microelectromechanical system equipped by a series of inertial measurement units (OPAL, ADPM Inc.) capable to extract – from the raw data acquired – ten posture and gait parameters. A Machine Learning (ML) approach, considering tree-based algorithms fed with the aforementioned features, was implemented in KNIME Analytics Platform (version 4.1.3, KNIME AG) to assess the degree of separability among the two classes of patients. Table 1 summarizes the results of the preliminary ML analysis conducted using three tree-based algorithms. This paper presents a potential strategy to distinguish quantitatively – through ML – patients affected by Ataxia and PD. Despite the unbalanced patients’ cohorts (PD subjects are double respect to Ataxia ones), ML tree-based algorithm demonstrated to discriminate the two classes achieving quite fair scores. Further research will involve a deeper evaluation of patients and additional/more informative kinematic parameters to better corroborate the preliminary results of this study.