Using Lower Limb Wearable Sensors to Identify Gait Modalities: A Machine-Learning-Based Approach

. 2023 Nov 17;23(22):9241.

doi: 10.3390/s23229241.

Affiliations

Free PMC article

Item in Clipboard

Liam David Hughes et al.

Sensors (Basel).

2023.

Free PMC article

Abstract

Real-world gait analysis can aid in clinical assessments and influence related interventions, free from the restrictions of a laboratory setting. Using individual accelerometers, we aimed to use a simple machine learning method to quantify the performance of the discrimination between three self-selected cyclical locomotion types using accelerometers placed at frequently referenced attachment locations. Thirty-five participants walked along a 10 m walkway at three different speeds. Triaxial accelerometers were attached to the sacrum, thighs and shanks. Slabs of magnitude, three-second-long accelerometer data were transformed into two-dimensional Fourier spectra. Principal component analysis was undertaken for data reduction and feature selection, followed by discriminant function analysis for classification. Accuracy was quantified by calculating scalar accounting for the distances between the three centroids and the scatter of each category’s cloud. The algorithm could successfully discriminate between gait modalities with 91% accuracy at the sacrum, 90% at the shanks and 87% at the thighs. Modalities were discriminated with high accuracy in all three sensor locations, where the most accurate location was the sacrum. Future research will focus on optimising the data processing of information from sensor locations that are advantageous for practical reasons, e.g., shank for prosthetic and orthotic devices.

Keywords:

accelerometer; discriminant function analysis; gait; machine learning; principal component analysis; wearable sensors.

PubMed Disclaimer

Conflict of interest statement

The authors declare no conflict of interest.

Figures

**Figure 1**

Diagram showing the experimental set-up and sensor attachment locations on the participants.

**Figure 2**

(a) Time course of a raw acceleration signal during a fast walking condition, showing six peaks in acceleration amplitude across the three-second period. (b) Spectrogram of the acceleration signal during a fast walking condition highlights six periods of large frequency bandwidth over three seconds. (c) Two-dimensional Fourier transform image of the fast walking, highlighting the spectrum of a repeated signal around two times per second.

**Figure 3**

(a) Time course of a raw acceleration signal during a fast walking condition showing six periods of peak amplitudes over three seconds. (b) A two–dimensional Fourier transform (2DFT) image of the fast walking waveform, highlighting a repeating waveform occurring more than two times per second. (c) Same as ‘a’ for normal walking, showing six periods of peak amplitudes over three seconds. (d) Same as ‘b’ for normal walking, with a decreased observed frequency of the spectral repetition highlighting a repeated waveform occurring roughly two times per second. (e) Same as ‘a’ for slow walking, showing five periods of peak amplitudes over three seconds. (f) Same as ‘b’ for slow walking, with a decreased observed frequency of the spectral repetition highlighting a repeated waveform occurring, highlighting a repeating waveform occurring under two times per second.

**Figure 4**

(a) Raw acceleration–time curves for one individual’s trial for sensors at the sacrum. The superimposed rectangular selections show three experimental conditions: normal, slow and fast walking. (b) A zoomed-in sacrum acceleration–time curve for fast walking. This shows the ten bouts of straight-line walking and the deceleration phases where the participant turned. These turns were dismissed from the analysis. (c) A stacked series of two-dimensional Fourier transform images displaying three experimental conditions. The superimposed rectangular selections show three experimental conditions: normal, slow and fast walking. Ten two-dimensional Fourier transform images were stacked per condition, which are three seconds in length, resulting in 10 spectra containing information on a three-second signal per experimental condition. Dark red indicates a higher acceleration magnitude, and dark blue indicates a lower magnitude. (d) A higher density series of two-dimensional Fourier transform images highlighting three separate experimental conditions. The superimposed rectangular selections show three experimental conditions: normal, slow and fast walking. Initial two-dimensional Fourier transform spectra are scanned temporally in time increments of 0.1 s over a period of one second, providing spectra for each experimental condition (100 × 3 s spectra per experimental condition).

**Figure 5**

Discriminant function scores show three activities clustered separately. Criterion scores were further calculated using the following equation: Discrimination criterion = $(\frac{Dist_1 \times Dist_2 \times Dist_3}{Scatter_01 \times Scatter_02 \times Scatter_03})$ . (a) Shows successful discrimination with a criterion score larger than 500, cloud centroids exhibit large separation between one another, and the cluster standard deviations are relatively small. (b) Shows unsuccessful discrimination with a criterion score smaller than 500, two experimental condition cloud centroids in close proximity (normal and fast walking), and relatively large cluster standard deviations.

**Figure 6**

Criterion scores for all participants (rearranged left-right from lowest to highest) for (a) sacrum, (b) left thigh, (c) right thigh, (d) left shank and (e) right shank sensors. (f) Highlights mean criterion scores for all sensors.

References

1. Rehman R.Z.U., Del Din S., Shi J.Q., Galna B., Lord S., Yarnall A.J., Guan Y., Rochester L. Comparison of Walking Protocols and Gait Assessment Systems for Machine Learning-Based Classification of Parkinson’s Disease. Sensors. 2019;19:5363. doi: 10.3390/s19245363.
  
  –
  
  DOI
  
  –
  
  PMC
  
  –
  
  PubMed
1. Buckley C., Alcock L., McArdle R., Rehman R., Del Din S., Mazzà C., Yarnall A., Rochester L. The Role of Movement Analysis in Diagnosing and Monitoring Neurodegenerative Conditions: Insights from Gait and Postural Control. Brain Sci. 2019;9:34. doi: 10.3390/brainsci9020034.
  
  –
  
  DOI
  
  –
  
  PMC
  
  –
  
  PubMed
1. Rehman R.Z.U., Zhou Y., Del Din S., Alcock L., Hansen C., Guan Y., Hortobágyi T., Maetzler W., Rochester L., Lamoth C.J.C. Gait Analysis with Wearables Can Accurately Classify Fallers from Non-Fallers: A Step toward Better Management of Neurological Disorders. Sensors. 2020;20:6992. doi: 10.3390/s20236992.
  
  –
  
  DOI
  
  –
  
  PMC
  
  –
  
  PubMed
1. Dauriac B., Bonnet X., Pillet H., Lavaste F. Estimation of the walking speed of individuals with transfemoral amputation from a single prosthetic shank-mounted IMU. Proc. Inst. Mech. Eng. H. 2019;233:931–937. doi: 10.1177/0954411919858468.
  
  –
  
  DOI
  
  –
  
  PubMed
1. Noh B., Yoon H., Youm C., Kim S., Lee M., Park H., Kim B., Choi H., Noh Y. Prediction of Decline in Global Cognitive Function Using Machine Learning with Feature Ranking of Gait and Physical Fitness Outcomes in Older Adults. Int. J. Environ. Res. Public Health. 2021;18:11347. doi: 10.3390/ijerph182111347.
  
  –
  
  DOI
  
  –
  
  PMC
  
  –
  
  PubMed