Validation of Low-Cost IMUs for Telerehabilitation Exercises

Abstract

Telerehabilitation, a specialized domain within telemedicine, supports remote physical rehabilitation and progress monitoring. Wearable sensors can improve this service by providing reliable monitoring of movement parameters, offering objective information into patients' rehabilitation sessions. This study presents the development and validation of a telerehabilitation system including a rehabilitation protocol, low-cost wearable inertial measurement units (IMUs) and a set of metrics descriptive of movement capacity to analyze rehabilitation exercises. Eleven medically stable elders (9 females, 2 males; age: 72.6 ± 5.0 years; height: 1.66 ± 0.09 m; mass: 67.8 ± 9.8 kg) performed 12 rehabilitation upper/lower limb and trunk exercises. Movement analysis was conducted using a prototypical IMU sensor and commercially available IMU as a reference. Each exercise was automatically segmented into single repetitions, from which selected metrics were computed. Bland-Altman analysis was performed to evaluate measurement agreement and consistency between the systems across all parameters. Results indicate acceptable measurement agreement for key rehabilitation metrics, including movement quantity, accelerations intensity, and movement smoothness. However, angular velocity and movement stability reveal technical limitations requiring refinement prior to clinical implementation. Balancing measurement reliability and affordability of telerehabilitation system remains a crucial factor to offer an effective service to individuals with diverse health conditions.

Keywords: IoT; inertial sensors; movement intensity; movement quality; physical exercises; telemedicine.

Figure 1

REHACT motor rehabilitation protocol. Lower limb exercises: KE1—knee extension lying down; KE2—knee extension sitting with a back support; KE3—knee extension without back support; SQ—half squat supported on a table. Upper limb exercises: SF1—shoulder flexion lying down; SF2—shoulder flexion sitting with back support; SF3—shoulder flexion without back support; PS—wall push-up. Trunk exercises: BD1—bird dog using only the legs; BD2—bird dog using only the arms; LF—lateral flexion of the column; RO—rotation of the column.

Figure 2

(a) Prototypical IMU, (b) positioning for the lower limb exercises, (c) positioning for trunk exercises, and (d) positioning for upper limb exercises.

Figure 3

Screenshot of the smartphone app to manage the IMUs.

Figure 4

Example of repetition identification in the prototype’s data. The principal axis (in this case, the X-axis), is segmented in single repetitions (vertical red dashed lines). In green, the used threshold.

Figure 5

Percentage values of prototype parameters (red line) are reported compared to the reference values (100% line reported in blue). The black dot represents the mean of the differences between the two devices (Bias), in percentage of the reference value. The dashed line highlights the amplitude of the limits of agreement (LoA), in percentage the reference value. Values on each radar are relative to Intensity and Quality parameters: peak acceleration measurements ($a_X^{peak}$, $a_Y^{peak}$, $a_Z^{peak}$), peak angular velocity (${\omega }_X^{peak}$, ${\omega }_Y^{peak}$, ${\omega }_Z^{peak}$), movement intensity and movement intensity variability (MI, MIV), range of angular velocity (RAV), log dimensionless jerk (LDLJ), and maximum value of range of motion (PKROM). Radars are provided for the following exercises: KE1—knee extension lying down; KE2—knee extension sitting with back support; KE3—knee extension without back support; SQ—half squat in support.

Figure 6

Percentage values of prototype parameters (red line) are reported compared to the reference values (100% line reported in blue). The black dot represents the mean of the differences between the two devices (Bias), in percentage of the reference value. The dashed line highlights the amplitude of the limits of agreement (LoA), in percentage the reference value. Values on each radar are relative to Intensity and Quality parameters: peak acceleration measurements ($a_X^{peak}$, $a_Y^{peak}$, $a_Z^{peak}$), peak angular velocity (${\omega }_X^{peak}$, ${\omega }_Y^{peak}$, ${\omega }_Z^{peak}$), movement intensity and movement intensity variability (MI, MIV), range of angular velocity (RAV), log dimensionless jerk (LDLJ), and maximum value of range of motion (PKROM). Radars are provided for the following exercises: SF1—shoulder flexion lying down; SF2—shoulder flexion sitting with back support; SF3—shoulder flexion without back support; PS—wall push-up.

Figure 7

Percentage values of prototype parameters (red line) are reported compared to the reference values (100% line reported in blue). The black dot represents the mean of the differences between the two devices (Bias), in percentage of the reference value. The dashed line highlights the amplitude of the limits of agreement (LoA), in percentage the reference value. Values on each radar are relative to Intensity and Quality parameters: peak acceleration measurements ($a_X^{peak}$, $a_Y^{peak}$, $a_Z^{peak}$), peak angular velocity (${\omega }_X^{peak}$, ${\omega }_Y^{peak}$, ${\omega }_Z^{peak}$), movement intensity and movement intensity variability (MI, MIV), range of angular velocity (RAV), log dimensionless jerk (LDLJ), and maximum value of range of motion (PKROM). Radars are provided for the following exercises: BD1—bird dog using only the legs; BD2—bird dog using only the arms; LF—lateral flexion of the column; RO—rotation of the column.