Validation and comparative study of the Motus system for accurately identifying movement behaviours using different sampling frequencies

Abstract

Abstract Accurate data on movement behaviours is essential for public health policy and interventions. Accelerometer-based tools like ActiPASS, paired with AxivityX3 accelerometers, are widely used but face challenges like slow data processing and administrative burden. To address these issues, we developed Motus (alias SurPASS), a scalable accelerometer-based system with wireless SENSmotionPlus accelerometers and cloud storage. While Motus shows promising feasibility, validating its performance at different sampling frequencies can further improve battery life, data transfer, and storage. This study aimed to validate Motus in identifying movement behaviours at different sampling frequencies and to compare it with ActiPASS. Data collection included laboratory and free-living sessions, during which participants wore three thigh-mounted accelerometers: SENSmotionPlus (25 Hz and 12.5 Hz) and AxivityX3 (25 Hz). The laboratory session was video recorded and comprised structured and semi-structured movement behaviours, including sedentary, standing, walking, stair climbing, running, and cycling. In the free-living session, participants wore accelerometers for 48 h and carried out usual activities. Data were processed using Motus, with free-living sessions additionally analysed with ActiPASS. The F1-score and balanced accuracy were used to assess Motus performance against video annotations, while Bland-Altman plots evaluated agreement between F1-scores (25 Hz vs. 12.5 Hz) and between Motus (25 Hz and 12.5 Hz) and ActiPASS, with ActiPASS as the reference method. Eighteen participants (61% female, age 34.1 ± 8.3 years) completed the data collection. A total of 395.7 min from the laboratory (mean: 22.0 ± 2.7 min) and 501,122 min from the free-living session (mean: 2638 ± 499.7 min) were analysed. In the laboratory, overall F1-score and balanced accuracy were 0.94 for both frequencies, compared with video observations. Additionally, mean bias F1-scores were ± 0.01 for all behaviours when comparing the performance of Motus 12.5 Hz vs. 25 Hz against videos. In free-living, Motus 25 Hz vs. ActiPASS showed a mean difference of ± 1 min for all behaviours. For Motus 12.5 Hz vs. ActiPASS, the mean difference were ± 1 min for sedentary, stair climbing and cycling, while a mean difference of + 5.1, -2.9 and − 2.2 min were observed for standing, walking and running, respectively. In conclusion, our study provides significant insights into the performance of Motus, highlighting the possibility of reducing the sampling frequency from 25 to 12.5 Hz and still obtaining high performance in classifying different movement behaviours.