Abstract

AZNIELLE, Tania Yadira  and  HERNANDEZ CACERES, Jose Luis. A Nonlinear Identification Approach for Gait Analysis of Amyotrophic Lateral SclerosisPatients. RCIM [online]. 2023, vol.15, n.1  Epub June 01, 2023. ISSN 1684-1859.

Gait data analysis, is giving mixing results regarding locomotion changes associated to Amyotrophic Lateral Sclerosis (ALS) development; the need has been claimed for new tools. We applied a nonlinear identification approach to the study of gait data from both healthy and ALS patients, available from Physionet.org. Kernel nonparametric nonlinear autoregression allowed to obtain noise-free realizations (NFR) that mimicked original traces, though correlation between original data and corresponding NFR was lower among ALS patients (p=0.03), suggesting a higher contribution of stochastic influences. Visual inspection of phase portraits, reconstructed from NFR via Takens theorem application, suggested dynamics differences between control subjects and patients. This was confirmed when phase portrait features were quantified and submitted to discriminant analysis (89% of correct classifications; 24/28). Application of a nonlinear dissimilarity measure for comparing pairs gait recordings, defined as a distance between underlying nonlinear autoregressive functions allowed an excellent separation between ALS and controls, via multidimensional scaling. Obtained projection map clearly suggested that ALS traces lay in a narrower dynamical space. This might reflect the known fact about neuronal degeneration accompanying ALS progression. When dissimilarity matrix principal components were introduced as predicting variables, discriminant analysis yielded an 82% of correct classifications (23/28). Overall, our results suggest that a nonlinear identification approach, centered in the characterization of the dynamics of the gait process can bring new insights to gait data interpretation.

Keywords : gait data analysis; non linear approach; Takens’ theorem.

        · abstract in Spanish     · text in English     · English ( pdf )