Anne-Marie Strauch

Advisors
Malte Ollenschläge (M.Sc.), Arne Küderle (M.Sc.), Dr. Felix Kluge, Prof. Dr. Björn Eskofier

Duration
05/2020-11/2020

Abstract
The way we walk can tell a lot about us. Gait analysis can be used for example in diagnosing pathologies [1], monitoring therapy success [2], or assessing fall risk [3]. However, gold standard methods for obtaining gait parameters, like step width or time, require expensive equipment and
are often constrained to laboratory conditions [4]. Body worn inertial measurement units (IMUs), which consist of accelerometers and gyroscopes, allow for ubiquitous gait analysis. They are cheap, unobtrusive, do not require laboratory conditions and therefore allow for the analysis of gait parameters in more situations and contexts [4]. Thus, they can complement more clinical approaches of disease diagnosis and monitoring treatments, like for example the Unified Parkinsons’s Disease Rating Scale [5, 6]. IMU-based gait parameters are typically obtained using double-integration [7] or complementary/ Kalman filters [8]. Some recent approaches also use neural networks [9]. These methods achieve a mean error in stride length estimation below one centimeter on average over multiple strides, but the standard deviation of this error is higher and seldom below six centimeters [9]. This means that IMU-derived gait parameters show a good validity for mean parameters, but poor validity for individual strides and measures of gait variability.

In contrast to mean parameters, the validity for variability parameters is less often reported and partly delivers contradicting results. The variability measures which are most often examined include standard deviation or root mean squares and the coefficient of variation calculated over
multiple consecutive strides. Rebula et al. reported that the root mean square of stride width and stride length were within 4% of the gold standard method [10]. In a study by Allseits et al. [11] the coefficient of variation of gait cycle time differs by 7% from the gold standard. The variability of single limb stance time shows a difference of 26% relative to the gold standard in the same study. Similarly, a further study found root-mean-squared coefficient of variation percentage between 31% and 56% for various variability parameters when compared to motion capture gold standard [12]. A possible reason for these errors in gait parameter calculation are random fluctuations in the angular velocity during swing phase [12]. However, no thorough analysis of possible reasons for high errors in gait parameter estimation has been conducted, especially not for foot-mounted IMUs.

We hypothesize that specic signal features can be found which predict under- or overestimation of gait parameters. Therefore, this work aims to investigate the correlation between the occurrence of specific signal features and errors in the stride length estimation regarding single
strides. Possibly this allows to identify the underlying physical reason for the occurrence of these signal features. Further on, this can help to tune algorithms to cope with the issue and to design sensors and sensor attachments that are better suited to prevent those errors.
In order to allow for a more in-depth analysis of one specic aspect in this field, this thesis will be limited to examining spatial gait parameters. The algorithm used for calculating these spatial parameters from raw sensor data is taken from [7] and the time stamps for stride segmentation are given as gold standard.

References:
[1] Gillain, S. et al.: The value of instrumental gait analysis in elderly healthy, MCI or Alzheimer’s disease subjects and a comparison with other clinical tests used in single and dual-task conditions. Annals of physical and rehabilitation medicine, 52 (6), 453-474, 2009.
[2] Block, J. A. and Shakoor, N.: The Biomechanics of Osteoarthritis: Implications for Therapy. Current rheumatology reports, 11 (1), 15-22, 2009.
[3] Rispens, S. M. et al.: Identification of Fall Risk Predictors in Daily Life Measurements: Gait Characteristics’ Reliability and Association With Self-reported Fall History. Neurorehabilitation and neural repair, 29 (1), 54-61, 2015.
[4] Muro-De-La-Herran, A. et al.: Gait analysis methods: An overview of wearable and non-wearable systems, highlighting clinical applications. Sensors, 14 (2), 3362-3394, 2014.
[5] Goetz, C. G. et al.: Movement Disorder Society-sponsored revision of the Unified Parkinson’s Disease Rating Scale (MDS-UPDRS): scale presentation and clinimetric testing results. Movement Disorders, 23 (15), 2129-2170, 2008.
[6] Hannink, J. et al.: Quantifying postural instability in Parkinsonian gait from inertial sensor data during standardised clinical gait tests. 2017 IEEE 14th International Conference on Wearable and Implantable Body Sensor Networks (BSN), 129-132, 2017.
[7] Rampp, A. et al.: Inertial sensor-based stride parameter calculation from gait sequences in geriatric patients. IEEE transactions on biomedical engineering, 62 (4), 1089-1097, 2014.
[8] Ferrari, A. et al.: A mobile Kalman-lter based solution for the real-time estimation of spatio-temporal gait parameters. IEEE transactions on neural systems and rehabilitation engineering, 24(7), 764-773, 2015.
[9] Hannink, J. et al.: Mobile Stride Length Estimation with Deep Convolutional Neural Networks. IEEE Journal of Biomedical and Health Informatics, 22 (2), 354-362, 2018.
[10] Rebula, R. J. et al.: Measurement of foot placement and its variability with inertial sensors. Gait & Posture, 38 (4), 974-980, 2013.
[11] Allseits, J. et al.: The development and concurrent validity of a real-time algorithm for temporal gait analysis using inertial measurement units. Journal of Biomechanics, 55, 27-33, 2017.
[12] Rantalainen, T. et al.: Reliability and concurrent validity of spatiotemporal stride characteristics measured with