Citation

Patashov D, Menahem Y, Ben-Haim O, Gazit E, Maidan I, Mirelman A, Sosnik R, Goldstein D, Hausdorff JM. Ann. Biomed. Eng. 2019; ePub(ePub): ePub.

Affiliation

Rush Alzheimer's Disease Center and Department of Orthopaedic Surgery, Rush University, Chicago, IL, USA.

Copyright

(Copyright © 2019, Holtzbrinck Springer Nature Publishing Group)

DOI

10.1007/s10439-019-02380-4

PMID

31598893

Abstract

In this study, we present algorithms developed for gait analysis, but suitable for many other signal processing tasks. A novel general-purpose algorithm for extremum estimation of quasi-periodic noisy signals is proposed. This algorithm is both flexible and robust, and allows custom adjustments to detect a predetermined wave pattern while being immune to signal noise and variability. A method for signal segmentation was also developed for analyzing kinematic data recorded while performing on obstacle avoidance task. The segmentation allows detecting preparation and recovery phases related to obstacle avoidance. A simple kernel-based clustering method was used for classification of unsupervised data containing features of steps within the walking trial and discriminating abnormal from regular steps. Moreover, a novel algorithm for missing data approximation and adaptive signal filtering is also presented. This algorithm allows restoring faulty data with high accuracy based on the surrounding information. In addition, a predictive machine learning technique is proposed for supervised multiclass labeling with non-standard label structure.

Language: en

Keywords

Adaptive filtering; Complex envelope; Dual multi-label forecasting; Kernel clustering; Missing data; Noisy signal; Peak detection; Quasi-periodic signal; Signal segmentation