Real-World Gait Detection Using a Wrist-Worn Inertial Sensor: Validation Study

Kluge, Felix; Brand, Yonatan E; Micó-Amigo, M Encarna; Bertuletti, Stefano; D'Ascanio, Ilaria; Gazit, Eran; Bonci, Tecla; Kirk, Cameron; Küderle, Arne; Palmerini, Luca; Paraschiv-Ionescu, Anisoara; Salis, Francesca; Soltani, Abolfazl; Ullrich, Martin; Alcock, Lisa; Aminian, Kamiar; Becker, Clemens; Brown, Philip; Buekers, Joren; Carsin, Anne-Elie; Caruso, Marco; Caulfield, Brian; Cereatti, Andrea; Chiari, Lorenzo; Echevarria, Carlos; Eskofier, Bjoern; Evers, Jordi; Garcia-Aymerich, Judith; Hache, Tilo; Hansen, Clint; Hausdorff, Jeffrey M; Hiden, Hugo; Hume, Emily; Keogh, Alison; Koch, Sarah; Mätzler, Walter; Megaritis, Dimitrios; Niessen, Martijn; Perlman, Or; Schwickert, Lars; Scott, Kirsty; Sharrack, Basil; Singleton, David; Vereijken, Beatrix; Vogiatzis, Ioannis; Yarnall, Alison; Rochester, Lynn; Mazzà, Claudia; Del Din, Silvia; Mueller, Arne

Wrist-worn inertial sensors are increasingly used to assess mobility in real-world settings. Reliable gait sequence (GS) detection is a prerequisite for estimating gait parameters, but remains challenging at the wrist due to complex arm movements. While gait detection algorithms exist for other sensor locations, validation at the wrist across diverse disease populations and comparisons with lower back sensors are lacking.
To validate wrist-based GS detection algorithms using real-world reference data and to compare their performance with algorithms applied to lower back–worn sensors.
Eighty-three participants with Parkinson disease, multiple sclerosis, proximal femoral fracture, chronic obstructive pulmonary disease, congestive heart failure, and healthy older adults were monitored for 2.5 hours in real-world conditions. Inertial sensors were worn on the wrist, lower back, and feet, with pressure insoles and infrared distance sensors serving as reference. Ten wrist-based GS detection algorithms were evaluated and compared with lower back–based approaches.
The best wrist-based algorithm achieved mean sensitivity ranging from 0.55 to 0.81 and specificity from 0.95 to 0.98 across disease groups. The mean relative absolute error of estimated walking time ranged from 8.9% to 32.7%. Performance was consistently lower than that of lower back–based algorithms, which showed higher sensitivity (0.71–0.91), specificity (0.96–0.99), and lower walking time error (6.3%–23.5%). Detection accuracy was reduced in participants with severe gait impairment and those using bilateral walking aids.
Wrist-based algorithms can reliably detect gait sequences in real-world settings and support extraction of gait parameters and daily walking patterns. However, lower back sensors provide superior performance. These findings support informed sensor placement choices in clinical and public health studies.

Supplementary resources

Rights

Use and reproduction:


CC BY 4.0

Please note that individual components of the publication may be subject to other licensing or copyright conditions.

Cite

Citation style:
Could not load citation form.