Analysis of the accuracy and reliability of a telerehabilitation system using a set of inertial measurement units on healthy subjects

According to the Italian national health service, stroke is one of the main causes of disability. After a stroke event the impaired patients need assistance and rehabilitation to restore their independence as much as possible. For mildly disability, rehabilitation consists of a first phase in the hospital and then at home where several patients drop out the therapy for lack of assistance. Recently, telerehabilitation has been playing a fundamental role, helping patients to continue their therapy. In this context, a home-based neuromotor telerehabilitation system (DoMoMEA) for mild-impaired stroke patients implementing a full-body rehabilitation protocol was presented to engage patients during the rehabilitation with motor exergames. The latter have been designed to monitor their progress over the weeks and give them motivational feedback. To this end, the real-time joint kinematics was estimated using a set of wearable inertial sensors (IMUs) and a set of biomechanical parameters such as the range of motion (ROM) was extracted. However, the accuracy of the joint kinematics time series and the reliability of the ROM could be affected by some factors like the anatomical calibration procedure or the errors due to the IMU orientation reconstruction. The aim of this work was to assess the accuracy of the joint kinematics estimation as provided by DoMoMEA system using the stereophotogrammetric system as reference. Moreover, the reliability of the estimated ROM was evaluated in terms of absolute agreement using the intraclass correlation coefficient (ICC) during a test-retest session of nine planar rehabilitation exercises including flex/extension of the elbow, wrist, knee, hip, ankle and trunk, ab/adduction of the shoulder, and the trunk rotation around both the vertical and the antero-posterior axes. Nine healthy subjects (25 ± 2.1 y.o.) were tested and equipped with eight wearable inertial measurement units (IMUs) applied with elastic straps on the body segments of interest and 49 retro-reflective markers (Davis protocol) for the reference joint kinematics. For the test session subjects performed 20 repetitions of each exercise at about 1/3, 2/3 and 3/3 of their maximum ROM and the values obtained with both systems were used to compute the root mean square difference (RMSD) between the time series. During retest subjects performed 20 repetitions of each exercise, at their maximum ROM and the twenty values were averaged to calculate the ICC(3,k). Average RMSD values of each exercise and different motion amplitudes were less than 6 deg for all joints but for the elbow (13 deg). These larger differences observed at the elbow joint are probably due to a different definition of anatomical upper arm axes for the two systems. The ICCs highlighted a good reliability of the IMU-based measurements for the exercises of the wrist, knee, ankle, and trunk rotation around antero-posterior axis with ICC between 0.79 and 0.89. A moderate reliability for elbow (0.64) and for shoulder, hip, and trunk flex/extension (0.69-0.73), probably due to the IMU’s elastic strap movement caused by muscle contractions, soft tissues wobbling, skin sliding. Then, a poor reliability resulted in the trunk rotation around the vertical axis (0.40) because the trunk movement was weakly constrained, and its amplitude was difficult to reproduce between sessions. Similar results were observed for the reference. The findings of this work are encouraging for future validation on stroke patients in a validation protocol.