Correction: Assessing real-world gait with digital technology? Validation, insights and recommendations from the Mobilise-D consortium
M. Encarna Micó-Amigo, Tecla Bonci, Anisoara Paraschiv-Ionescu, Martin Ullrich, Cameron Kirk, Abolfazl Soltani, Arne Küderle, Eran Gazit, Francesca Salis, Lisa Alcock, Kamiar Aminian, Clemens Becker, Stefano Bertuletti, Philip M. Brown, Ellen Buckley, Alma Cantu, Anne‐Elie Carsin, Marco Caruso, Brian Caulfield, Andrea Cereatti, Lorenzo Chiari, Ilaria D’Ascanio, Bjoern M. Eskofier, Sara Johansson Fernstad, Marcel Froehlich, Judith García‐Aymerich, Clint Hansen, Jeffrey M. Hausdorff, Hugo Hiden, Emily Hume, Alison Keogh, Felix Kluge, Sarah Koch, Walter Maetzler, Dimitrios Megaritis, Arne Mueller, Martijn Niessen, Luca Palmerini, Lars Schwickert, Kirsty Scott, Basil Sharrack, Henrik Sillén, David Singleton, Beatrix Vereijken, Ioannis Vogiatzis, Alison J. Yarnall, Lynn Rochester, Claudia Mazzà, Silvia Del Din, Sara JohanssonJournal of NeuroEngineering and Rehabilitation2024 · DOI · OpenAlex
Digital mobility outcomes (DMOs) calculated from wearable devices require rigorous technical validation to be used as reliable clinical endpoints. This paper provides a formal correction to the technical specifications of gait sequence detection (GSD) algorithms within the Mobilise-D framework. The consortium assessed DMOs—including GSD, initial contact detection (ICD), cadence (CAD), and stride length (SL)—across six clinical cohorts: healthy older adults, Parkinson’s disease, multiple sclerosis, proximal femoral fracture, chronic obstructive pulmonary disease, and congestive heart failure. Results identified cohort-specific top-performing algorithms, particularly for GSD and CAD. While ICD and CAD algorithms showed excellent performance (relative errors < 11%), SL estimation remains more challenging. This correction ensures the reproducibility of the GSD algorithms, which are vital for monitoring patients with significant mobility impairments in real-world settings.
Introduction: Addressing the Reproducibility Gap
In large-scale clinical studies using wearable sensors, reproducibility is paramount. This update corrects the technical blueprint for the Mobilise-D gait analysis pipeline, specifically addressing errors in the documentation of GSD algorithms. Beyond correcting typographical errors, this update clarifies how algorithmic choices—such as moving from single-axis analysis to the acceleration norm—mitigate practical challenges like sensor misalignment in unsupervised environments. In the "real world," patients may wear devices slightly tilted or even upside down; the framework must be robust enough to handle this "noise" without losing data integrity.
Methods: The GSD Algorithmic Framework
The Mobilise-D framework utilizes three primary GSD algorithms to identify walking bouts from raw sensor data collected via a single wearable device worn on the lower back. The validation study compared these against a reference system combining inertial modules, distance sensors, and pressure insoles.
As detailed in the corrected Table 1, the GSD algorithms are categorized as follows:
$GSD_A$ (Frequency-based): Employs band-pass filtering (0.5–3 Hz) and convolution with a 2 Hz sine wave template to define gait regions.
$GSD_B$ (Time-domain): Recommended for impaired cohorts, it calculates the acceleration norm (vector magnitude of all three axes), applies a low-pass FIR filter ($f_c = 3.2$ Hz), and uses a peak detection threshold of 0.1g.
$GSD_C$ (Time-domain): Follows the same approach as $GSD_B$ but utilizes a higher peak detection threshold of 0.15g.
The shift to the acceleration norm makes the system "axis-agnostic," ensuring that gait detection remains robust regardless of minor sensor orientation changes.
Key Results and Technical Specifications
The correction establishes definitive parameters for implementation, ensuring that researchers can replicate the consortium's findings:
Filter Specifications: Time-domain methods ($GSD_B$ and $GSD_C$) must use an FIR filter with a 3.2 Hz cutoff.
Thresholding: The primary difference between the time-domain variants is the peak detection threshold (0.1g vs 0.15g).
Performance Metrics: Best GSD algorithms showed good performance with sensitivity > 0.73 and accuracy > 0.94. ICD and CAD algorithms presented even stronger results, with relative errors under 11% and 8.5%, respectively.
Output Format: Detected sequences are output as a $1 \times N$ vector of start and end times (in seconds) relative to the recording start.
Discussion: Robustness in the Real World
The use of the acceleration norm is a critical design choice for real-world applications. By capturing the overall magnitude of movement rather than relying on a perfectly aligned vertical axis, $GSD_B$ provides superior performance for cohorts with slow gait speeds and irregular patterns, such as those recovering from hip fractures.
Furthermore, the inclusion of an adaptive step duration threshold allows the system to "learn" individual rhythms. This is essential for heterogeneous clinical populations where gait may be unstable or highly variable. However, it is important to note that performance across all DMOs was lower for the cohort with the most severe impairments (proximal femoral fracture) and for very short walking bouts or slow gait speeds (< 0.5 m/s).
Conclusion
The identified and corrected algorithms enable robust estimation of key digital mobility outcomes. While environmental factors like walking bout length and speed influence performance, the cohort-specific application of these validated algorithms provides a reliable foundation for digital gait assessment in clinical trials and long-term patient monitoring.
Summary updated
Abstract
Background: Although digital mobility outcomes (DMOs) can be readily calculated from real-world data collected with wearable devices and ad-hoc algorithms, technical validation is still required. The aim of this paper is to comparatively assess and validate DMOs estimated using real-world gait data from six different cohorts, focusing on gait sequence detection, foot initial contact detection (ICD), cadence (CAD) and stride length (SL) estimates. Methods: Twenty healthy older adults, 20 people with Parkinson’s disease, 20 with multiple sclerosis, 19 with proximal femoral fracture, 17 with chronic obstructive pulmonary disease and 12 with congestive heart failure were monitored for 2.5 h in the real-world, using a single wearable device worn on the lower back. A reference system combining inertial modules with distance sensors and pressure insoles was used for comparison of DMOs from the single wearable device. We assessed and validated three algorithms for gait sequence detection, four for ICD, three for CAD and four for SL by concurrently comparing their performances (e.g., accuracy, specificity, sensitivity, absolute and relative errors). Additionally, the effects of walking bout (WB) speed and duration on algorithm performance were investigated. Results: We identified two cohort-specific top performing algorithms for gait sequence detection and CAD, and a single best for ICD and SL. Best gait sequence detection algorithms showed good performances (sensitivity > 0.73, positive predictive values > 0.75, specificity > 0.95, accuracy > 0.94). ICD and CAD algorithms presented excellent results, with sensitivity > 0.79, positive predictive values > 0.89 and relative errors < 11% for ICD and < 8.5% for CAD. The best identified SL algorithm showed lower performances than other DMOs (absolute error < 0.21 m). Lower performances across all DMOs were found for the cohort with most severe gait impairments (proximal femoral fracture). Algorithms’ performances were lower for short walking bouts; slower gait speeds (< 0.5 m/s) resulted in reduced performance of the CAD and SL algorithms. Conclusions: Overall, the identified algorithms enabled a robust estimation of key DMOs. Our findings showed that the choice of algorithm for estimation of gait sequence detection and CAD should be cohort-specific (e.g., slow walkers and with gait impairments). Short walking bout length and slow walking speed worsened algorithms’ performances. Trial registration ISRCTN – 12246987.
BibTeX
@article{micoamigo2024correction,
author = {M. Encarna Micó-Amigo and Tecla Bonci and Anisoara Paraschiv-Ionescu and Martin Ullrich and Cameron Kirk and Abolfazl Soltani and Arne Küderle and Eran Gazit and Francesca Salis and Lisa Alcock and Kamiar Aminian and Clemens Becker and Stefano Bertuletti and Philip M. Brown and Ellen Buckley and Alma Cantu and Anne‐Elie Carsin and Marco Caruso and Brian Caulfield and Andrea Cereatti and Lorenzo Chiari and Ilaria D’Ascanio and Bjoern M. Eskofier and Sara Johansson Fernstad and Marcel Froehlich and Judith García‐Aymerich and Clint Hansen and Jeffrey M. Hausdorff and Hugo Hiden and Emily Hume and Alison Keogh and Felix Kluge and Sarah Koch and Walter Maetzler and Dimitrios Megaritis and Arne Mueller and Martijn Niessen and Luca Palmerini and Lars Schwickert and Kirsty Scott and Basil Sharrack and Henrik Sillén and David Singleton and Beatrix Vereijken and Ioannis Vogiatzis and Alison J. Yarnall and Lynn Rochester and Claudia Mazzà and Silvia Del Din and Bjoern Eskofier and Sara Johansson},
title = {{Correction: Assessing real-world gait with digital technology? Validation, insights and recommendations from the Mobilise-D consortium}},
journal = {Journal of NeuroEngineering and Rehabilitation},
year = {2024},
doi = {10.1186/s12984-024-01361-6},
}