Ecological validity of a deep learning algorithm to detect gait events from real-life walking bouts in mobility-limiting diseases

• DOI: 10.3389/fneur.2023.1247532
• Publication Date: 2023-10-16T16:30:16Z
• Authors: Robbin Romijnders; Francesca Salis; Clint Hansen; Arne Küderle; Anisoara Paraschi...; Andrea Cereatti; Lisa Alcock; Kamiar Aminian; Clemens Becker; Stefano Bertuletti; Tecla Bonci; Philip Brown; Ellen Buckley; Alma Cantu; Anne-Elie Carsin; Marco Caruso; Brian Caulfield; Lorenzo Chiari; Ilaria D'Ascanio; Silvia Del Din; Björn Eskofier; Sara Johansson Fe...; Marceli Stanislaw...; Judith Garcia Aym...; Eran Gazit; Jeffrey M. Hausdorff; Hugo Hiden; Emily Hume; Alison Keogh; Cameron Kirk; Felix Kluge; Sarah Koch; Claudia Mazzà; Dimitrios Megaritis; Encarna Micó-Amigo; Arne Müller; Luca Palmerini; Lynn Rochester; Lars Schwickert; Kirsty Scott; Basil Sharrack; David Singleton; Abolfazl Soltani; Martin Ullrich; Beatrix Vereijken; Ioannis Vogiatzis; Alison Yarnall; Gerhard Schmidt; Walter Maetzler
• Keywords: Limiting; Ecological validity

Introduction The clinical assessment of mobility, and walking specifically, is still mainly based on functional tests that lack ecological validity. Thanks to inertial measurement units (IMUs), gait analysis shifting unsupervised monitoring in naturalistic unconstrained settings. However, the extraction clinically relevant parameters from IMU data often depends heuristics-based algorithms rely empirically determined thresholds. These were validated small cohorts supervised Methods Here, a deep learning (DL) algorithm was developed for event detection heterogeneous population different mobility-limiting disease cohort healthy adults. Participants wore pressure insoles IMUs both feet 2.5 h their habitual environment. raw accelerometer gyroscope used as input convolutional neural network, while reference timings events combined data. Results discussion results showed high-detection performance initial contacts (ICs) (recall: 98%, precision: 96%) final (FCs) 99%, 94%) maximum median time error −0.02 s ICs 0.03 FCs. Subsequently derived temporal good agreement with insoles-based mean difference 0.07, −0.07, <0.01 stance, swing, stride time, respectively. Thus, DL considered successful detecting ecologically valid environments across diseases.