Monitoring motor capacity changes of children during rehabilitation using body-worn sensors

Abstract

Background: Rehabilitation services use outcome measures to track motor performance of their patients over time. State-of-the-art approaches use mainly patients' feedback and experts' observations for this purpose. We aim at continuously monitoring children in daily life and assessing normal activities to close the gap between movements done as instructed by caregivers and natural movements during daily life. To investigate the applicability of body-worn sensors for motor assessment in children, we investigated changes in movement capacity during defined motor tasks longitudinally.

Methods: We performed a longitudinal study over four weeks with 4 children (2 girls; 2 diagnosed with Cerebral Palsy and 2 with stroke, on average 10.5 years old) undergoing rehabilitation. Every week, the children performed 10 predefined motor tasks. Capacity in terms of quality and quantity was assessed by experts and movement was monitored using 10 ETH Orientation Sensors (ETHOS), a small and unobtrusive inertial measurement unit. Features such as smoothness of movement were calculated from the sensor data and a regression was used to estimate the capacity from the features and their relation to clinical data. Therefore, the target and features were normalized to range from 0 to 1.

Results: We achieved a mean RMS-error of 0.15 and a mean correlation value of 0.86 (p < 0.05 for all tasks) between our regression estimate of motor task capacity and experts' ratings across all tasks. We identified the most important features and were able to reduce the sensor setup from 10 to 3 sensors. We investigated features that provided a good estimate of the motor capacity independently of the task performed, e.g. smoothness of the movement.

Conclusions: We found that children's task capacity can be assessed from wearable sensors and that some of the calculated features provide a good estimate of movement capacity over different tasks. This indicates the potential of using the sensors in daily life, when little or no information on the task performed is available. For the assessment, the use of three sensors on both wrists and the hip suffices. With the developed algorithms, we plan to assess children's motor performance in daily life with a follow-up study.

Figures

Figure 1

Child wearing sensors and close ups of the sensors.a) Child wearing 10 ETHOS units (highlighted with circles; bracelet (b) and flat (c) housing types were used) during walking with a walking frame.

Figure 2

Pictures of tasks performed during weekly sessions.

Figure 3

Schematic overview of dataset and procedure of data analysis.

Figure 4

Acceleration signal of the right wrist during the cards task. The standard deviation of the acceleration magnitude was calculated with a 0.5 s sliding window with a 0.49 s overlap to estimate when the hand was moved and when not. The detected on- and off-set of movement were then used to estimate task completion time.

Figure 5

Acceleration of the right (upper figure) and left (lower figure) wrist of subject 1, week 1, performing the pick up small objects task. Note that using the left hand, the subject took twice as long to complete the task. While the pick up of the individual 6 objects can be identified for the left hand (indicated with boxes), this is more challenging for the right hand since the movement is faster and includes more dynamic movement as opposed to measuring mainly acceleration due to gravity. The MI and MIV features can be extracted from the histograms depicted on the right side of the figure: MI is indicated with the mean acceleration value and MIV can be associated with the spread of the histogram. The spread and thus MIV are larger for the unaffected (right) hand.

Figure 6

Energy spectrum obtained via the Fourier Transform of the left (affected) and right (unaffected) hand performing the task “pick up small objects and place in bin” of subject 1. It can be observed that the energy associated to the dominant frequency of the affected hand was much lower than that of the unaffected hand. We used both the dominant frequency and the energy associated to it as features. The SM parameter of the unaffected side was almost twice as high as that of the affected side.

Figure 7

Right (solid) and left (dashed) wrist acceleration during wheel chair driving of subject 1. Push cycles are marked. Each push cycle consists of a forward pushing phase (a) and a phase when the hands are moved back to prepare for the next push (b). It can be observed that synchrony is high for the forward push (a) and lower for the backward release (b). The left hand moves back later, which is the affected hand of this subject.

Figure 8

Calculated stance duration of the left and right foot during the TUG test across the four weeks of subject 2. It can be seen that the rating increased while the ratio of left to right stance duration approximated 1, the movement thus became less asymmetric.

Figure 9

Motor function rating by movement scientists and as estimated by our regression model. Model includes data from the three subjects who were able to walk across the four weeks. The first two subjects increased their performance over the four weeks.

Figure 10

Ground truth and regression estimation of the motor function using sensor data. Only statistically significant features (p<0.05, α=0.05) were used for the estimation, see Table 4. Note that for the single-handed tasks the results for the left and right hand performing the task are depicted whereas for the bimanual tasks there is only one set of graphs. Subplots are ordered with decreasing RMSE. The best result was achieved for the Turn Around Cards and the NHPT tasks.

Figure 11

Scatter plot of the three best task-independent features identified during the data analysis, namely average rotation energy (ARE), range of angular rotation (RANG), and smoothness of movement (SM). Points indicate the feature values of the different subjects and weeks averaged over all tasks. Points are colored according to the performance rating of the expert. It can be observed that the points with better ratings have higher RANG, higher ARE, and higher SM. Since these values are the average over all performed tasks, this figure indicates the potential of these features for performance assessment in daily life when little or no information on the performed task is available.