Ataxia and Parkinson's disease patients classification using tree-based machine learning algorithms fed by spatiotemporal features: a pilot study

Ataxia and Parkinson's Disease (PD) are two of the neurodegenerative diseases more studied in recent years. To identify and/or monitor the progression of such pathologies, several quantitative gait analysis investigations have been performed. Albeit several publications have presented and discussed the results of this strategy in case of such pathologies (considered individually and/or compared to healthy subjects), to the best of the authors' knowledge two or more pathologies have been scarcely investigated simultaneously from the point of view of quantitative gait analysis. To this aim, the main objective of this paper is to presents a new strategy to classify and/or monitor the disease progression of Ataxia and PD using quantitative gait analysis tools and, additionally, machine learning strategies. Specifically, a group of 43 and 22 PD and ataxic patients, respectively, was enrolled and subjected to a quantitative gait analysis investigation - using a microelectromechanical system (the OPAL) - as a result of which ten spatiotemporal quantitative features were automatically extracted. These parameters were fed to five tree-based machine learning algorithms which demonstrated to successfully distinguish Ataxia (following an oversampling step) and PD patients, showing promising metrics, e.g., accuracies and sensitivities up to 81% and 86% The findings obtained suggest the proposed strategy could drive new research questions in the field and be of direct practical relevance in the clinical setting.