Timing estimation for gait in water from inertial sensor measurements: Analysis of the performance of 17 algorithms. (December 2020)

• Record Type: Journal Article
• Authors: Pacini Panebianco, Giulia; Bisi, Maria Cristina; Stagni, Rita; Fantozzi, Silvia
• Is Part Of: Computer methods and programs in biomedicine. Volume 197(2020)
• Publication Date: 2020-12
• Subjects: Algorithm -- Event detection -- Gait events -- Gait timing -- IMUs -- Temporal parameters -- Walking -- Water -- Wearable inertial sensors; WW Walking in water -- WDL Walking on dry land -- GE Gait Event -- GTP Gait Temporal Parameter -- FC Foot Contact -- FO Foot Off -- IMU Inertial Measurement Unit -- FIR Finite Impulse Response -- IIR Infinite Impulse Response -- WT Wavelet Transform -- GRF Ground Reaction Force -- E Error -- PPV positive predictive value -- Med Median -- Dmed Dispersion around median value -- ICC Intraclass correlation coefficient -- CV Coefficient of Variation; Medicine -- Computer programs -- Periodicals; Biology -- Computer programs -- Periodicals; Computers -- Periodicals; Medicine -- Periodicals; Médecine -- Logiciels -- Périodiques; Biologie -- Logiciels -- Périodiques; Biology -- Computer programs; Medicine -- Computer programs; Periodicals; Electronic journals; 610.28
• DOI: 10.1016/j.cmpb.2020.105703
• Languages: English
• ISSNs: 0169-2607

Abstract

Highlights: Timing estimation with inertial sensors was assessed for gait in water environment. Performance of 17 algorithms was analysed based on implementation characteristics. No trunk-based algorithm is suitable for gait timing estimation in water. Shank/foot-based algorithm performance depends on analysed target variable. Target variable ICC supports the assessment of algorithm performance in altered gait. Abstract: Background and objectives: Walking in water is used for rehabilitation in different pathological conditions. For the characterization of gait alterations related to pathology, gait timing assessment is of primary importance. With the widespread use of inertial sensors, several algorithms have been proposed for gait timing estimation (i.e. gait events and temporal parameters) out of the water, while an assessment of their performance for walking in water is still missing. The purpose of the present study was to assess the performance in the temporal segmentation for gait in water of 17 algorithms proposed in the literature. Methods: Ten healthy volunteers mounting 5 tri-axial inertial sensors (trunk, shanks and feet) walked on dry land and in water. Seventeen different algorithms were implemented and classified based on: 1) sensor position, 2) target variable, and 3) computational approach. Gait events identified from synchronized video recordings were assumed as reference. Temporal parameters were calculated from gait events. Algorithm performance was analysed in terms of sensitivity, positive predictive value, accuracy, and repeatability. Results: For walking in water, all Trunk-based algorithms provided a sensitivity lower than 81% and a positive predictive value lower than 94%, as well as acceleration-based algorithms, independently from sensor location, with the exception of two Shank-based ones. Drop in algorithm sensitivity and positive predictive value was associated to significant differences in the stride pattern of the specific analysed variables during walking in water as compared to walking on dry land, as shown by the intraclass correlation coefficient. When using Shank- or Foot-based algorithms, gait events resulted delayed, but the delay was compensated in the estimate of Stride and Step time; a general underestimation of Stance- and overestimation of Swing-time was observed, with minor exceptions. Conclusion: Sensor position, target variable and computational approach determined different error distributions for different gait events and temporal parameters for walking in water. This work supports an evidence-based selection of the most appropriate algorithm for gait timing estimation for walking in water as related to the specific application, and provides relevant information for the design of new algorithms for the specific motor task.