All sensors were worn over clothing using a lightweight Velcro elasticated webbing system for keeping the sensors in place. All participants wore standardized running shoes Lonsdale of the correct shoe size, in order to correct for differences in mobility due to shoe stiffness or heels; our team have a collection of different sizes of these running shoes to fit all participants.
Sensors were placed on the lateral surfaces of thighs, to avoid interference with walking; sensors were orientated with the positive X-axis pointing superiorly proximally. These six scales plus the demographics scale required approximately 20 min to fill in. Fitting the sensors took 5 min, while removing the sensors took 3 min. In general the entire procedure for a single volunteer lasted 60 min including rest time.
The sensors had their data synchronized at the beginning and the end of the experiment by being affixed together and being subjected to sudden transient accelerations, interspersed with periods of non-movement. The timed-up-and-go TUG task was performed according to Steffen et al. The task involves arising from a seated position, walking 3 m, turning around, walking back 3 m, turning around and sitting back down in the chair. Participants started in a chair with arms, with a tape mark on the floor showing the 3 m distance where they were supposed to turn around.
The TUG task was performed twice. Participants were also asked to perform several other walking and balance tasks, including a TimedFoot-Walk [T25FW based on timing with a stopwatch, 46 ], which was used to establish that participants were at the Hauser Ambulation Index [HAI, 47 ] of 2 or below. None of the tasks were stressful or tiring, and participants were asked before each task if they needed a rest. These sensors are factory calibrated for gravitational acceleration accelerometers and angular momentum gyroscopes , and they incorporate an onboard algorithm for estimation of heading and quaternions 48 , These sensors have been validated for accuracy when measuring walking, both in terms of angular velocities and derived temporal gait metrics Time alignments between sensors and with other measurements and video tapes were performed using an automated event-based synchronization strategy [e.
Directions used i. Binary file sensor data was transferred to a Windows 7 computer, and the binary files were converted into csv files using the manufacturer's provided Graphical User Interface.
The csv files were read into Matlab, and all sensor data was aligned based on the synchronization signals at the beginning and end of the experiment with a purposed-made script; timing differences between sensors were interpolated linearly—at no point did the original sensor acquisition data differ between sensors by more than 50 ms over the course of 90 min of acquisition.
The relevant sensor data for each task was located by Matlab based on the event's start and finish time recorded by the sensor, and all data was low-pass filtered 2. Timing duration from the spine sensor was based on Weiss et al. To test control of movement, repeated gait movements can be tested for variation, such as the Coefficient of Variation for any metric e. For a single movement performed once e. In this study, we tested two different measures of smoothness.
The normalized mean absolute jerk 54 is one of the most commonly used measures for smoothness smoothness 1 :. Another measure of smoothness we used, the speed arc length 55 , has the advantage of being unit-free smoothness 2 :. To allow for peaks from different legs and in different directions to be compared, all peaks are the peak of the absolute value of the calibrated signal, and all means are also the mean of the absolute value of the calibrated signal.
Graphical inspection of healthy and PwMS peak angular velocity data showed that it was approximately normally distributed; nevertheless, to allow for those features that were not normally distributed, for assessments of correlation between repeated attempts of the same task, an Intraclass Correlation Coefficient ICC was calculated For unpaired comparisons between the means of two populations, the Wilcoxon Rank Sum test was used; this was corrected by the Holm-Bonferroni correction for multiple comparisons.
For effect size calculations, the rank biserial was calculated. The two cohorts compared in the main study were ambulatory persons with multiple sclerosis PwMS and middle-aged healthy volunteers.
The baseline characteristics of the two groups are shown in Table 1. Pitch gyroscope data from each sensor and roll data from the lumbar spine sensor were used to derive both the rate of movement during the sit-to-stand and stand-to-sit transitions, as well as the durations that these activities lasted.
Representative sensor data is shown in Figure 3. All traces in this figure are low pass filtered 2. Panels A healthy and D PwMS show both left and right thigh pitch traces during the entire TUG task; each walking step is clearly identifiable from the swing phase sharp peaks and concurrent contra-lateral stance phase wider, blunt peaks , as are the sit-to-stand and stand-to-sit transitions wider and lower-amplitude changes.
The turns are more easily identified by the traces for the yaw gyroscopes not shown. Figure 3. Representative traces of pitch gyroscope sensors data during the TUG task.
A shows the activity of the left red and right dark blue thigh sensors during the entire TUG task for a healthy volunteer. B,C show expanded views of the same representative traces at the sit-to-stand transition B and the stand-to-sit transition C , labeled with key points for feature calculation.
D—F show analogous traces for a PwMS; note that the different panels have slightly different scales on their axes. In addition to the pitch traces from the left thigh red and the right thigh dark blue , B,C,E,F include a pitch trace from the lumbar spine sensor black , to allow comparisons with previously published data features based on torso-mounted sensor data. The start of the rise for the left thigh is a red circle, for the right thigh is a dark blue circle, and for the spine is a magenta diamond.
Figure 3B is a close up of panel A during the sit-to-stand transition showing the relationship between the peaks of the spine pitch trace black line and the thigh traces. For the purposes of computer identification, zero-crossing points of the thigh traces black squares were used as markers for the end of SI-ST transitions. Panel C is a close up of panel A during the stand-to-sit transition showing the relationship between the peaks of the spine pitch trace and the thigh traces; for this volunteer, the second spine peak right-most vertical gray line is closely aligned with the thighs' return to the seat pan of the chair i.
The delay of the thigh pitch traces red and blue traces, between The first spine peak is delayed compared to knee and thigh flexion cyan circle on red line at 9.
The thigh activity of the right lower limb dark blue is a combination of the final shuffling step during Turn 2 T2, starting at the dark blue circle and the subsequent flexion of sitting down. The traces related to a PwMS in panel D show a similar set of activities as in panel A, although the actions are performed more slowly and with lower angular velocity peaks.
The most noticeable difference is that in panel F the ST-SI transition is performed much more slowly and carefully. Figure 4 shows a close up view of the same left thigh pitch trace during the sit-to-stand transition from Figures 3A,B , along with the peak attributes and time points used to derive the features for these movements. A complete description of the arcs is provided in the Supplementary Materials.
Arcs A-H correspond to the sit-to-stand transition, while arcs J-R correspond to the same attributes during the stand-to-sit transition there is no arc I. Arcs E and N not shown correspond to a 1-s time period centered around the maximum i. The peak shown here as a black circle is bracketed by the step end to the right, black square and the start of the rise to the left, dark blue triangle.
Figure 4. Arc boundaries used for calculations of features. The left thigh pitch trace from the sit-to-stand transition in Figure 3 is labeled with the relevant time markers and peak attributes used to calculate the features in this study. How these points were computationally derived is described in the methods; note that arcs E and N not shown are 1 s regions centered on the peak, and arc I does not exist. Before determining which features were most likely to be affected in our cohort by MS, we sought to determine which of the features were clearly repeatable.
Because each of the participants performed the TUG task twice, we compared the value of each feature during the first attempt and the second attempt. The features we tested were based on the pitch angular velocity measurements from both thighs and the spine sensor, roll angular velocity measurements from the spine sensor, a range of smoothness metrics, and an omnibus measure of TUG duration based on the Anterior-Posterior accelerometer of the thigh.
The calculations were the absolute value magnitude of the peak angular velocity, the many possible durations of the event as determined by the arcs as explained in the methods and Figure 4 , the magnitude of the mean angular velocities for those arcs, the area under the curve for those arcs, and the smoothness of each arc see Methods. Each pitch feature was initially calculated for both left and right thighs and also for the spine , and the final thigh features were the maximum of the two thigh values, the minimum of the two thigh values, the value associated with the thigh making the first step, and the value associated with the thigh making the second step.
Representative plots showing selected correlations of four of the features are shown in Figure 5. The most correlated measurement arcs for the transitions are arc J, K, N, and M all of which encompass the entirety of the ST-SI peak including the peak itself ; the least correlated were arcs P and Q, both of which represent the first half of the ST-SI transition.
The vast majority of smoothness metrics were poorly correlated, although a few were good between 0. This may be expected, given that lack of smoothness would represent loss of control, which would per force be inconsistent.
Figure 5. Persons with MS are shown as red circles, healthy volunteers are shown as blue triangles. A shows excellent correlation between the two TUG trials each participant performed for the feature: the absolute value of the peak pitch angular velocity during the Stand-to-Sit ST-SI transition magenta circle in Figure 3C. C shows the absolute value of the mean angular velocity of the signal as shown as arc F in Figure 4 during the Sit-to-Stand transition.
D shows the correlations for the duration of the sit-to-stand phase arc F. In total correlated features were tested, and they were compared between the healthy volunteers and the PwMS. Those not included in the table were redundant or similar to other features already in the table [e. To account for multiple comparisons, the Holm-Bonferroni method was used. To illustrate the scale of those differences, a comparison of the total TUG task durations as measured by stopwatch are shown next to this plot see Figure 6B.
Figure 6. Comparison of TUG variables for Healthy vs. Black horizontal lines are mean values. When comparing the effect sizes Rank Biserial in Table 2 of MS in our cohort of the features, several observations arise. The features relating to the sit-to-stand transition have a larger effect size and are more consistently relevant when discriminating PwMS from healthy volunteers than the stand-to-sit transition.
The angular velocity features Area under the Curve, absolute peak and absolute mean have larger effect sizes and are more consistently relevant when observing PwMS than the durations.
In our hands, the effect sizes of the durations arising from the spine sensor [features 21 and 22 in this study, originally from Weiss et al. As stated above, smoothness features were less consistent than other features. As an unplanned analysis, we sought to understand how these variables might work together, given that many of the features were based on similar or related measurements.
Using a stepwise procedure Matlab , we removed variables that were weak contributors low absolute t -values or were not robust when subsets of volunteers were selected for the model. A set of seven features were found and described in a logistic regression see Table 3. The regression had an R 2 [coefficient of discrimination 59 ] of 0.
None of the pairs of variables had a coefficient of correlation above 0. To check for overfitting, combined data for healthy and PwMS volunteers were randomly split in half training set , betas were re-derived for the seven robust features, and the remaining volunteers test set were compared to predicted values based on the new betas; in attempts, the average correct prediction rate was 0.
Inertial sensor metrics of gait and mobility variables, and their responsiveness to clinical conditions, are being explored for the differences elicited by sensor placement on different parts of the body In this study of MS, we considered myriad TUG features derived from previous studies of ambulatory disabilities of all kinds , and found informative metrics derived from thigh-positioned wearable inertial sensors that would be useful for estimating disability in PwMS, particularly with regard to strength and effort.
We also compared a range of the best of the thigh-based metrics to spine-based metrics which represent both strength and control , and ran a logistic regression on the results.
We list seven non-overlapping features that may be useful together as complementary metrics in assessments of disability progression in MS, and also as metrics for clinical efficacy for interventions proposed to improve or limit disability in MS. Our novel contribution is to consider the combination of thigh and spine metrics in MS—as did Motta et al.
Our data specifically considers the case of TUG, which includes the SI-ST and ST-SI transitions; these transitions are particularly challenging activities in everyday life, and are especially revealing of the movement of the thigh segment. In our cohorts we compared a wide variety of sensor-based micro-features of TUG to two timing features of TUG as a whole; we found that many of the thigh-derived sensor micro-features are reproducible and have high reliability, and that a collection of thigh pitch angular velocity features including absolute values of the area under the curve, the peak and the mean based on the sit-to-stand transition differed between MS and healthy with higher effect sizes rank biserial than total time duration of TUG; three of these features were statistically significantly different between healthy and PwMS by the stringent Holm-Bonferroni method of multiple comparisons.
These features were all similar measurements of the area under the curve for pitch angular velocity for the SI-ST transition. Because the SI-ST transition is a demanding task for the musculature, and higher values for pitch angular velocity would be particularly demanding, we associate these variables with strength This fits with previous research on patients with total knee arthroplasty that concluded that quadriceps weakness has a substantial impact on performance of the sit-to-stand task 20 , We also tested temporal duration features based on the thigh SI-ST transition and previously published features based on the spine-derived SI-ST transition 52 , and we found the set of such spine-derived features that were potentially useful, but those features resulted in lower effect sizes than the traditional stopwatch duration of TUG for our cohorts and thus had lower effect sizes than the best angular velocity features.
For both sit-to-stand and stand-to-sit transitions, spine data is discriminatory, but thigh data is more discriminatory for MS disability. We also measured many features suggesting that thigh pitch or spine pitch is much more discriminatory than spine roll.
In a logistic regression we found that our initial hypothesis was supported: the movement of the thigh during the SI-ST transition was the most informative of all the TUG measures tested, and that adding a thigh feature feature 3 robustly improved a logistic regression compared to using only spine features with the total TUG duration.
However, we were surprised to find that five of the seven robust features were from the spine sensor, three were related to roll, and two were related to smoothness; none of the other thigh features were independent or robust enough to stay in the analysis after the first one was included. Of the spine features, it is intuitive that healthy volunteers have a large pitch SI-ST peak feature 14, implying torso strength and effort , and that PwMS have a larger roll peak during ST-SI feature 26, implying loss of torso control.
It also makes some sense that healthy volunteers would have a smoother roll in angular momentum in the 1 s surrounding the ST-SI peak feature 27, arc N, Figure 4. It is less intuitive that the spine roll signal during most of the SI-ST transition feature 22, arc D would be smoother for MS patients than for healthy volunteers; presumably this relates to MS patients being slower and more cautious when rising using the chair's arms , but none of the other calculations peak, mean or duration is discriminatory in this way.
A previous study examining the shank-mounted sensor metrics of TUG as an entire task in PwMS 16 found that their regression models for clinical disability metrics [EDSS and Multiple Sclerosis Impact Scale MSIS ] incorporated many sensor metrics of angular velocity including mean angular velocities, maximum angular velocities, and minimum i. In a study of the elderly 33 , 52 , the range of the vertical accelerometry signal located at the lumbar spine was a discriminatory feature for identifying idiopathic fallers among the elderly, while SI-ST duration and ST-SI duration were not discriminatory.
The use of inertial sensor technology in clinical assessment of disability is moving ahead rapidly in both MS and in disorders of mobility more generally. The goal of such systems is to increase the resolution and consistency of measurements of ambulatory disability e. EDSS 4. Only further sensor research on clinical populations will clarify whether this goal is even possible. Currently a commercial system for measuring mobility during TUG that is operated by clinicians i.
Extensive research into this particular inertial sensor methodology has been driven by the manufacturer of this system, which places sensors near the ankles. He has covered almost every collegiate sport imaginable, including March Madness on a few occasions. By: Brian Polking. Published: 18 December, More Articles. Home Yoga Stretching. Disinfect, wash and dry your laboratory glassware. Contact us today to find out more about our range. Servicing scientific laboratories since For Accuracy and Professionalism.
Suppliers of chemicals, consumables and scientific laboratory equipment. Read More ». HandyStep Touch S. The next generation in repetitive pipetting. New and improved models for , in stock and available for immediate delivery at Rowe Scientific.
Intuitive User Interface UI. The new and improved intuitive touchscreen UI offers expanded functionality via the clearly structured graphical menus.
0コメント