Stroke Rehabilitation (II): Robotic Assessment of Proprioception

Previous studies have found the importance of somatosensory signals in motor control. Deaffarented patients are shown to have unique and distinct movement patterns compared to healthy control. Within the scope of motor behavior, movement-related somatosenses here means bodily senses that cover proprioception (sense of joint position), kinesthesia (sense of limb movement), and other senses originating from the mechanoreceptors, but does not include pain and thermal receptors. A large literature also shows that the somatosensory system is important for motor learning.

Neurological injuries (e.g. stroke, Parkinson's Disease, and multiple sclerosis) are often accompanied by somatosensory deficits. There has been a growing interest to assess and train the somatosensory system in a clinical setting. In the case of stroke, the somatosensory test is part of a standard clinical assessment. However, a recent systematic review (Connell & Tyson, 2012) has shown that such evaluation is known to be unreliable. The authors commented that the sensory section of the Fugl-Meyer Assessment and Erasmus version of Nottingham Sensory Assessment has the best balance between usability and robustness. In a stroke study, it was found that proprioception and stereognosis were more frequently impaired than tactile sensations and that there is little agreement in impairment among different somatosensory modalities (Connell et al, 2008). More research on assessments and interventions for somatosensory impairment post-stroke is warranted.

Broadly speaking, there are two classes in the assessment of proprioception: joint-position matching (JPM) tasks and psychophysical threshold method (PTM).

JPM tasks evaluate the ability to replicate the position or velocity of a joint angle or a limb position in the absence of vision. Participants respond by replicating the reference movement or position using the same limb (unilateral protocol) or the contralateral limb (bilateral). 

On the other hand, PTM tasks assess the ability to detect the intensity of proprioceptive stimuli, or discriminate two stimuli equal in nature but differ in amplitude/intensity. For example, the stimuli can be tactile, position, angle, or vibration. Participants respond typically with a verbal response.

Although some traditional tools such as goniometers can be used, the reliability of such measurements is unproven. For this reason, the use of robotic devices in the assessment of somatosensory integrity becomes popular. Robotics promise increased precision and accuracy, in addition to better reliability compared with standardized observer-based ordinal scales. Examples of such devices include KINARM (bimanual upper limb), MIT Manus (upper limb), Lokomat (lower limb), Wristbot (wrist), AMADEO (fingers). Other devices include an isokinetic dynamometer for the lower limb (Biodex).

Joint Position Matching Task
Joint position matching (JPM) tasks can be exercised on both lower and upper limbs, and either using unilateral or bilateral matching limbs. A thorough review of this test is described by Goble (2010). Instead of a pure sensory test, JPM is seen to involve some working memory component. Let's begin with the lower limb assessment. A few studies used Biodex lower limb devices to study lower limb proprioception. In a study by Willem et al (2002), subjects lied down in a supine position with the ankle in position 15° plantar-flexion. Active and passive joint-position sense using JPM method was assessed at the ankle, and muscle strength (isokinetic peak torque) was determined. This is also an example of a unilateral protocol, i.e. the same limb is used to replicate the movement.

Fig-1: An example of the joint position matching task using the same limb of lower extremities.

Another study investigates the hip and knee joint proprioception in healthy subjects and incomplete SCI subjects (Domingo and Lam, 2014) using a Lokomat exoskeleton. During the task, the subject's knee (or hip) was first moved passively by the robot into a position (joint angle θ). After a short break, the robot then moves it to a so-called distractor position. The subject was then asked to bring the joint back to the original position by using a joystick control (θ_S). The nominal value |θ ─ θ_S| is the matching error, reflecting the subject's performance. Refer to the figure above.

For the upper limb, studies mostly involved the more proximal limb, i.e. shoulder and elbow joints. A study with a bimanual protocol or two-arm matching task used two groups of subjects. In the first group, subjects moved both middle fingers to the same spatial location (extrinsic). The second group, however, was asked to move the right finger to a mirror-symmetric location of the left middle finger with respect to the body midline (intrinsic). The authors showed that bimanual accuracy is higher for tasks involving extrinsic coordinate (Iandolo, et al, 2015).

KINARM has proven to be a popular robotic tool for bimanual tasks studying sensorimotor behavior in the healthy and clinical population. Dukelow et al did a series of studies using KINARM in patients with stroke and see whether proprioception can be assessed objectively. In one study, 74 healthy subjects and 113 subjects with acute stroke (62 left-affected, 51 right-affected) performed a JPM task with vision occluded. The robot moved the most affected arm at a preset speed of 28 cm/sec, direction, and magnitude. Stroke patients were then asked to mirror-match the movement with their opposite, active arm (Semrau et al, 2013). For control subjects, both arms were tested. All subjects performed 36 total movements, for a total of 6 movements in each of 6 movement directions. See the figure below. The authors found that most stroke patients (69% of left-affected; median, 28.0°; 49% of right-affected subjects; median, 22.1°) made significantly larger directional errors than 95% of the controls (95% of range, <22.4°; median, 14.8°).

Fig-2: An example of the joint position matching task using bilateral limbs of upper extremities.

Recently, Konczak, Masia, and colleagues (2016) used a wrist robotic device and found the anisotropic nature of sensory acuity using JPM method on the dominant arm. Active matching acuity for flexion/extension was found to be 4.64 ± 0.24°; abduction/adduction: 3.68 ± 0.32°; supination/pronation: 5.15 ± 0.37°. A similar study was also performed with young children (Marini F, et al., 2017) and it was found that the anisotropic nature of sensory acuity does not change significantly across age. See the figure below for more illustration.

Fig-3: An example of the joint position matching task of the wrist (distal upper limb).

Psychophysical Threshold Method

Assessment using PTM yields a certain psychometric function with nominal value or threshold of detection in movement speed or position/angle. The method utilizes principles of Psychophysics, which is discussed in a separate blog entry on the "Human Sensory Perception". The shape of this function reflects the variability of responses about this threshold. The most common paradigm used in PTM tasks is a method of constant stimuli but one drawback of such method is the length procedure (Simo et al, 2014); and the same trial usually repeats until 3 to 5 correct judgments of the same stimulus are achieved. More recently, a revised and faster version of PTM paradigm was proposed by Mrotek et al (2017) using the "method of adjustment" during ten iterative trials. Here, subjects repeatedly adjust the magnitude of the stimulus until it is just perceptible, with an equal number of trials approaching that estimated threshold from below (i.e., starting from smaller stimulus magnitudes) and above (starting from larger magnitudes). Typical force magnitude and the threshold of detection can be found below for a hemiparetic patient (left) and normal control (right).

Fig-4: An illustration of a psychophysical method in assessing the threshold of detection.

Studies involving the lower limb have investigated the use of robotic devices in assessing knee joint proprioception. Using a custom-made device similar to the one by Biodex, Hurkmans et al. (2007) assessed the smallest detectable angular change in the knee joint. A similar method was used to find the smallest detectable passive knee movement in both sagittal as well as the frontal plane (Cammarata, et al., 2011). This method lets the subjects respond when they are just able to detect a change in joint position. In another study by Lam at UBC (Chrishold et al, 2016) in patients with spinal cord injury, Lokomat was used to test hip and joint again. Here, the joint was moved passively in 4 different movement speeds (0.5, 1.0, 2.0, and 4.0 deg/s) in both flexion and extension directions. The subject had to respond when the movement was first felt.

Assessment of proprioception in conjunction with motor adaptation has been studied by Ostry et al (2010). This study involves MIT-Manus robotic arm that produces a velocity-dependent curl field which perturbs the movement of the upper limb. Sensory acuity was assessed using an adaptive staircase method or PEST. The authors found that motor adaptation causes a shift in sensory acuity, a term called sensory recalibration (by another group, Henriques et al.). In another study, threshold detection of a wrist movement was assessed using a Wristbot by Konczak and colleagues. The authors found that the mean threshold for wrist flexion was 2.15°± 0.43° and 1.52°± 0.36° for abduction.

Conclusion
The robotic-based somatosensory assessment has gained popularity. It has great potential clinical applications in areas that involve human movements and movement disorders. Until now, there is no clinical biomarkers to predict somatosensory impairment in stroke (see: Boyd et al, 2018 for a review).


Reference: Based on a nice book chapter, "Robotic techniques for the assessment of proprioceptive deficits and for proprioceptive training" by Casadio et al., 2018. Figures are taken from the relevant individual citation therein.