In this proof of concept study, we used an exoskeleton-based assessment protocol to demonstrate the convergent validity of the acquired kinematic data in comparison with the UE-FMA clinical outcome measure in severely impaired stroke patients. The characteristics of the assessment protocol were as follows:
- It provides real-time movement feedback of the measurement system’s seven degrees of freedom by virtue of a biomorphic virtual representation of the upper limb, including the proximal component of the upper limb.
- It adjusts the device position and length of the forearm, upper arm and wrist appliances to the individual anatomical proportions to align posture and reduce patient movement within the exoskeleton.
- It applies the same position (neutral zero) with a distance of 90 degrees between forearm and upper arm as the starting position for all assessments to minimize the exoskeleton-patient interaction.
- It assesses, separately and subsequently as self-contained tasks, the range of motion/spatial posture of single joints (i.e., joints’ angles) and the closing and opening of the hand with a pressure sensor placed in the handle.
The majority of stroke patients will presumably never regain full function of the affected limb [13, 14]. The impairment of an upper extremity is a decisive factor for their diminished quality of life [15]. Early and high-dose movement therapies are therefore relevant for clinically meaningful improvements [16]. Furthermore, an assessment of upper limb movements is crucial in monitoring and understanding sensorimotor recovery [17].
Assistive technologies supporting robot-assisted and gravity-balancing training are therefore now under investigation as a means of increasing and standardizing the scope of therapy and assessment for stroke rehabilitation. These devices facilitate intensive, high-dose and repetitive exercises of activities of daily living (ADL), and may also support the therapists by an objective assessment of the upper limb. In this context, the gravity-balancing support by a seven degree-of-freedom arm exoskeleton may be particularly useful during the rehabilitation training and assessment of severely affected stroke patients [18, 19]. Together with other assessment options, this arm exoskeleton, a version of which is now commercially available, is suitable for real-time recording of wrist/arm/shoulder movements and finger strength.
In this context, extended custom-made soft- and hardware applications have been implemented and investigated to augment the functionalities: Online feedback of extent of movement and quality for the assisted ADL-like exercises [7, 8], closed-loop task difficulty adaptation of these virtual reach-to-grasp tasks [9], and hybrid exoskeletons including adaptive neuromuscular stimulation [20], additional brain control [21] and robotic support with active actuation [22].
It is, however, important to determine the convergent validity of these exoskeleton-based assessments, i.e., the degree to which the correlation of metrics to clinical scales is consistent with the hypothesis [1]. The Fugl-Meyer assessment [5] is still the gold standard and is commonly used for clinical movement assessment in hemiparetic stroke patients. Excellent interrater and intrarater reliability have been demonstrated, and the score is responsive to changes in motor impairment [23]. This assessment is, however, known to have its limitations. One study suggests that the sensory scale’s psychometric properties should not be used to assess chronic stroke patients [24]. The pain domain could be a confounding variable to the joint range of motion, which would render it unnecessary. Moreover, distal fine motor functions may be underrepresented. In addition, a ceiling effect of the motor function has been reported [25]. In this context, provided that they show convergent validity with the UE-FMA, the use of exoskeleton-based measurements may allow for a finer and more specific registration of movements [3].
Furthermore, the UE-FMA is relatively time-consuming and cannot be performed in the context of an exercise session, e.g., to track the improvement during a training period. With the assessment task presented here, detailed information about the progress on the impairment level can be provided during each training unit with minimal additional time expenditure. Importantly, this information is not confounded, as the task is self-contained and specified for the evaluation of motor impairment [1]. In particular, this assessment task does not include those training tasks applied during exoskeleton-based therapy, and the results pertaining to upper limb function are, therefore, not confounded by task-specific learning effects [6].
Our setup contained an integrated virtual reality module to provide immediate and continuous feedback of the movement extent. Such approaches are vital to motor learning in rehabilitation [7, 9, 26] and have been expanded here to the area of instrumental assessment. In this context, virtual reality is often used to provide feedback and might be beneficial during rehabilitation [8, 27-29]. To build an interactive interface for such purposes, movement data from the patients is acquired by various mechanical or optical systems, e.g., CyberGlove [30], orthotic exoskeletons [18], gaming systems [31-33], or in combination with robotic systems for haptic feedback such as Rutgers Master II-ND haptic glove, MIT-Manus [34] or ARMIN [35]. Devices such as the Armeo Spring, ARMIN [35], Pneu-Wrex [36], ULEXO7 [37], have the advantage of providing at least partial kinematic registration of the upper extremity movement for different joints. By contrast, systems such as the MIT-Manus [34], ReaPLAN [38, 39], ReoGo [40], Planar robot [40], and PUPArm [41] allow for endpoint-based alignment. With these devices, the movement of the shoulder and upper arm is estimated as a surrogate parameter, and not directly via sensors. The algorithm proposed in this study can be easily transferred to those former systems that allowed the registration of the different joints. This approach would, therefore, provide a standardized assessment that could be performed on different devices and enable better comparisons of the studies.
We assessed all single-joint angles in a standardized fashion that is often applied in the clinical context of neurological and orthopedic evaluations (i.e., beginning from the normal zero position with 90 degrees distance between forearm and upper arm). This distinguishes our approach from previous methods for kinematic movement assessment that were acquired during specific exercises with more complex movements [42]. Furthermore, various biomechanical parameters are often used for classification such as inter-joint coordination [43, 44], temporal efficiency [45-47], movement speed [48-50], smoothness [51] or accuracy [52]. Such classification algorithms tend to be rather device-specific and are not easily transferable to other hard- and software settings. Furthermore, sometimes even elaborate models are necessary, using, for example, artificial neuronal networks, to generate clinically relevant parameters [53]. On the whole, a large number of the identified kinematic parameters were derived from existing exercises within individual systems and compared with the UE-FMA score. Range-of-motion-based evaluation systems, for example, are available for task-related assessments [54, 55]. However, only very little data is available concerning single-joint-based movement assessment of, for example, wrist motion [56] or elbow movement [57]; moreover, no significant results are available for these approaches. Only when shoulder rotation was included, were significant correlations with clinical scores reported [58, 59]; this was also the case when compensatory movement patterns, which contribute to pathological upper arm coordination, were investigated [52, 60].
The choice of task, measurement system and metrics for the presented upper limb kinematic assessment addressed first and foremost the impairment level of severely affected patients who were unable to perform these movements without gravity-balancing. We demonstrated that shoulder rotation, a frequently neglected proximal component of kinematic assessments of the upper limb [3], was the measure with the most relevant contribution to the prediction of the clinical outcome measure. This may be related to the characteristics of the applied clinical outcome measure for stroke patients, i.e., the UE-FMA scale. The over-representation of the shoulder in the different evaluation tasks, a known feature of the UE-FMA, contributes to the overall score [25].
The second-best parameter that predicted the clinical status was, notably, the grip force, which tends to be under-represented in the UE-FMA. This finding may suggest that this fairly straightforward instrumental measure – which can be easily acquired even without an exoskeleton – may be best suited to a practical quantification of clinically relevant kinematics in a wide variety of patients after stroke. However, future work would need to study larger and independent sample sizes to confirm the predictive properties of these kinematic parameters. Furthermore, as there is no benchmark to compare the range of motion measured with the exoskeleton, in future work, participants may perform the task twice with two different operators to provide a measure of repeatability.