Automatic stride interval extraction from long, highly variable and noisy gait timing signals Tom Chau a,b,c, * , Sidra Rizvi a,b a Bloorview MacMillan ChildrenÕs Centre, 350 Rumsey Road, Toronto, Ont., Canada M4G 1R8 b Institute of Biomaterials and Biomedical Engineering, 4 Taddle Creek Road, University of Toronto, Toronto, Canada M5S 3G9 c Graduate Department of Rehabilitation Science, 256 McCaul St., Toronto, Canada M5T 1W5 Abstract This paper presents a probabilistic algorithm for automatically extracting the stride interval time series from long, highly variable and noisy two-state timing signals. Long interstride tem- poral records are of particular interest in nonlinear dynamical analysis of gait. The proposed method consists of probabilistic estimation and extraction followed by post-extraction filter- ing. With noisy timing signals from 10 children with Spastic Diplegia, no statistical differences in the numbers of extracted strides (p ¼ 0:94), the mean stride intervals (p ¼ 0:55) and the scal- ing exponents (p ¼ 0:94) (a measure of temporal heterogeneity) were found between series ex- tracted by hand and by the probabilistic algorithm. The method is robust to noise and violations of normality. Results support the use of probabilistic extraction as an alternative to laborious manual extraction. Ó 2002 Elsevier Science B.V. All rights reserved. PsycINFO classification: 2240 Keywords: Gait dynamics; Quasi-periodic events; Gait timing signals; Automatic extraction Human Movement Science 21 (2002) 495–514 www.elsevier.com/locate/humov * Corresponding author. Present address: Bloorview MacMillan ChildrenÕs Centre, 350 Rumsey Road, Toronto, Ont., Canada M4G 1R8. Tel.: +1-416-425-6220x3515; fax: +1-416-425-1634. E-mail address: ttkchau@ieee.org (T. Chau). 0167-9457/02/$ - see front matter Ó 2002 Elsevier Science B.V. All rights reserved. PII:S0167-9457(02)00125-2