Machine learning applied to the analysis of the gait initiation process in Parkinson’s disease using Principal Component Analysis and bootstrapping
Résumé
The experimental characterization of the parallel nature of gait initiation (GI) is usually done through the simultaneous acquisition of about 25 variables, resulting in very large datasets, and is performed using univariate analysis (the distribution of each variable of the dataset is explored one at a time). Considering the complexity of the physiological processes involved in the GI, we applied Principal Component Analysis (PCA) in order to identify the most meaningful sets of variables and their correlations. This multivariate statistical analysis is very powerful at reducing the dimensionality of large dataset and emphasizing correlations between variables. The bootstrapping technique, which consists in resampling the dataset with replacement, was also successfully applied in order to enhance the precision on the estimated parameters. Since Parkinson is one of the major neurodegenerative disease, those methods were applied to identify the 3 majors sets of variables influencing the postural control disability of Parkinsonian patients during GI. Finally, the significant benefit of a rehabilitation intervention consisting in the prior stretching of the Triceps Surae is shown (Fig1).