Fig 1.
BiostampRC sensors are skin-mounted, conformal devices that can be adhered at multiple locations on the body.
(A) BioStampRC skin-mounted, conformal motion sensor. (B) Anatomical locations where devices were adhered to the skin.
Table 1.
Error metrics for treadmill walking data from healthy subjects for 7 device location combinations.
Fig 2.
Performance of model using sacrum, thigh, and shank device locations on treadmill data from healthy subjects.
Bland-Altman (A) and regression (B) plots illustrating the performance of the accelerometer-based model for estimating walking speed on a treadmill. As illustrated in the plots, the model produces unbiased estimates of speed with homoscedastic error.
Table 2.
Error metrics for treadmill 6MWT data from MS patients for 7 device location combinations.
Fig 3.
Performance of model using sacrum, thigh, and shank device locations on treadmill data from MS patients.
Bland-Altman (A) and regression (B) plots illustrating the performance of the accelerometer-based model for estimating walking speed in MS patients. As illustrated in the plots, the model produces unbiased estimates of speed with slightly higher variance at slower speeds, despite being trained on data from subjects with normal gait.
Table 3.
Error metrics by MS impairment groups and Pearson product moment correlation between error and EDSSSR and MSWS scores.
Fig 4.
Scatter plots showing the relationship between speed estimation error and MSWS and EDSSSR.
Speed estimation error vs. MSWS score (A) and EDSSSR score (B) from the sacrum, thigh, shank model. The dashed red line is a line of best fit, correlations between error and EDSSSR/MSWS scores are not significant.
Table 4.
Relationship between walking speed and MSWS score, EDSSSR score, and fall history.