Kinematic Evaluation via Inertial Measurement Unit Associated with Upper Extremity Motor Function in Subacute Stroke: A Cross-Sectional Study

Kinematic evaluation via portable sensor system has been increasingly applied in neurological sciences and clinical practice. However, conventional kinematic evaluation rarely extends the context beyond the motor impairment level. In addition, kinematic tasks with numerous items could be complex and time consuming that pose a burden to test applications and data processing. The study aimed to explore the correlation of finger-to-nose task (FNT) kinematics via Inertial Measurement Unit with upper limb motor function in subacute stroke. In this study, six FNT kinematic variables were used to measure movement time, smoothness, and velocity in 37 participants with subacute stroke. Upper limb motor function was evaluated with the Fugl-Meyer Assessment for Upper Extremity (FMA-UE), Action Research Arm Test (ARAT), and modified Barthel Index (MBI). As a result, mean velocity, peak velocity, and the number of movement units were associated with the clinical assessments. The multivariable linear regression models could estimate 55%, 51%, and 32% of variance in FMA-UE, ARAT, and MBI, respectively. In addition, age, gender, type of stroke, and paretic side had no significant effects on these associations. Results show that FNT kinematic variables measured via Inertial Measurement Unit are associated with upper extremity motor function in individuals with subacute stroke. The objective kinematic evaluation may be suitable for predicting clinical measures of motor impairment and capacity to understand upper extremity motor recovery and clinical decision making after stroke. This trial is registered with ChiCTR1900026656.


Introduction
Upper extremity (UE) motor function is impaired in approximately 50-80% of individuals with acute stroke [1] and 40-50% with chronic stroke [2,3]. Motor impairment results in poor movement control and has a major impact on functional capacity and activities of daily living (ADL) of stroke survivors [4]. To optimize UE recovery after stroke, it is crucial to select multilevel outcome measures for the interpretation of motor recovery and clinical decision making [5]. Although there have been extensive validated UE scales or tests to assess body structure, function, and activity in clinical practice [6], these assessments often rely on subjectively rated ordinal scales with ceiling effects that may lead to examiner bias or lack sensitivity to detect potentially impactful changes of upper limb motor recovery [7].
Kinematic evaluation facilitates interpreting the mechanisms of motor restoration, which has been increasingly applied in neurological sciences and clinical practice [8][9][10]. Such technology is capable of providing detailed information regarding upper extremity function evaluation and delivering personalized interventions. According to the previous literature, kinematic assessment is usually performed using arm-supported robots or optical-camera systems, based on fixed laboratory environments or expensive equipment that leads to several disadvantages [11][12][13]. From the technical perspective, robotic instruments are unable to capture the entire spectrum of UE motor impairment due to their mechanical structure [14]. Moreover, most robotic devices could not extend the value of kinematic scenarios beyond the impairment level according to the International Classification of Functioning, Disability and Health (ICF) framework [11,15]. Optical camera systems raise privacy concerns inevitably and constrain participants into a laboratory environment with much setup time and cost [16].
Portable sensor systems, a novel approach of kinematic evaluation, can provide upper limb spatiotemporal measurements against gravity in a natural three-dimensional environment [17]. Inertial Measurement Units (IMU) are portable sensor devices combining the three-dimensional accelerometers, gyroscopes, and magnetometers to detect kinematic parameters. Kinematic analysis of motor impairment via Inertial Measurement Unit has been shown to be objective, sensitive, and quantitative. However, kinematic tasks with numerous items could be complex and time consuming that pose burden on test application, compliance issues, and data processing in previous studies [18,19]. Moreover, its relationship with the multilevel UE clinical measures regarding ICF framework has not been fully investigated [14,20].
In clinical practice, the finger-to-nose test (FNT) is commonly applied to evaluate upper limb coordination in patients with stroke and cerebellar ataxia [12,21]. Compared with multi-item clinical scales that require trained personnel and as long as 30 minutes to complete, FNT could reduce task burden when estimating individual's UE performance [22]. Previous studies have shown that FNT could add value to measure UE coordination with construct, convergent, and discriminant validity [12,23] as well as ADL-related dexterity [24,25]. However, it remains unclear how FNT correlates with motor impairment, capacity, and ADL performance in individuals with subacute stroke from a kinematic perspective. erefore, the purpose of this study was to explore the associations between FNT kinematic variables obtained via Inertial Measurement Unit and multilevel upper extremity motor function in subacute stroke survivors. Furthermore, we aimed to compare the amount of variance in clinical scales that could be explained by FNT kinematic variables. Hypothetically, kinematic metrics reflecting the UE movement strategy, smoothness, and velocity could be considered to measure more aspects of motor impairment (FMA-UE) than activity assessments (ARAT and MBI).

Study Design.
is cross-sectional study followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) checklist.
e study was performed in accordance with the principles of the Declaration of Helsinki. e study protocol was approved by the Clinical Trials Ethics Committee of Huazhong University of Science and Technology on 24 October 2018. e study was registered in the Chinese Clinical Trial Registry (no. ChiCTR1900026656) on 17 October 2019.

Participants.
irty-seven individuals with subacute stroke were recruited from the Department of Rehabilitation Medicine from December 2019 to January 2021 ( Figure 1). e inclusion criteria were as follows: (a) clinical diagnosis of unilateral, first-ever subacute stroke verified by MRI or CT; (b) aged 18-80 years; (c) showing upper limb motor impairment (Fugl-Meyer Assessment of Upper Extremity <66); (d) able to complete the kinematic protocol; (e) no complicating medical history, such as visual, cardiac, or pulmonary disorders. ose who had other musculoskeletal or neurological conditions that affected arm function were excluded from the study [23]. All the participants were right handed [26] and have provided written informed consent prior to study entrance.

Clinical Assessments. Clinical assessments of the participants included the Fugl-Meyer Assessment of Upper
Extremity (FMA-UE), Action Research Arm Test (ARAT), and modified Barthel Index (MBI). e FMA-UE is a validated and reliable assessment of poststroke upper limb motor impairment. FMA-UE is composed of 33 items that comprise four subscales (arm, wrist, hand, and coordination) regarding motor domains, and higher scores indicate less motor impairment of upper extremity [27]. e ARAT was used to evaluate UE functional capacity, including grasp, grip, pinch, and gross movement. ARAT consists of 19 fourpoint ordinal items, and higher scores indicate greater arm functional capacity [28]. e independence level in basic activities of living was assessed with the MBI, which consists of 10 items and higher scores indicate greater ADL independence [29].

Kinematic Assessment.
Kinematic assessment was implemented with an Inertial Measurement Unit system (IMU, Noraxon USA Inc.). Each IMU sensor contains a coordinate system to assess accelerations and three-dimensional orientations at a 100 Hz sampling frequency. e IMU system had shown excellent reliability, accuracy, and precision in quantifying kinematic test [17]. Four sensors were placed on body segments, including head, upper arm, forearm, and hand, to detect UE kinematic information. Participants were required to sit in a height-adjustable chair with their hips and knees flexed to 90°. Upper extremity maintained in the neutral position, with elbow extension and palm downward initially. Standardized procedure for the finger-to-nose test was first presented by the same researcher, and then, it was imitated by the participants for three times before testing. e tests were recorded for five times, and a mean of the variable was used in statistical calculations [30].
Data were extracted through a semiautomated code in MATLAB software ( e MathWorks, Natick, Massachusetts, USA) according to the anatomical coordinate system and joint rotation recommended by the International Society Biomechanical (ISB) [31]. Onsets and ends of FNT movements were defined with a velocity threshold of 50 mm/s [30]. In this cross-sectional study, six FNT kinematic variables were calculated: movement time (MT), mean velocity (VM), peak velocity (VP), percentage of time to peak velocity (TVP%), number of movement units (NMU), and normalized integrated jerk (NIJ) [30,32]. MT was an objective quantitative variable defined as the time spent during the test to reflect movement performance. e maximum tangential velocity of index finger was calculated during each movement segment to get VP; VM was defined as the average tangential velocity. TVP% was the proportion of time taken from the onset of the movement to the peak velocity. e number of velocity peaks exceeding 10% of VP was characterized as NMU. NIJ was utilized to assess movement smoothness, which was calculated using jerk, M,T and length of the task according to the following formula: where jerk represented the third derivate of end point displacement, and length represented the shortest distance between initial and terminal positions of index finger.

Statistical Analysis.
Statistical analysis was performed on IBM Statistical Package for Social Science (SPSS) version 22.0. Chi-squared test was used to examine categorical variables, and one-way ANOVA was used to examine quantitative variables. Shapiro-Wilk test or Q-Q plot was used to evaluate whether the quantitative data were normally distributed. Pearson's correlation coefficients (r) were conducted between kinematic and clinical assessments. e limit for multicollinearity among independent variables was set at 0.7 for correlation coefficients. After controlling the influencing factors (including age, gender, type of stroke, and paretic side), the kinematic metrics were included as independent variables into the multivariable linear regression to investigate the associations with clinical assessments. Probability for entry in backward regression was set at 0.05 and removal at 0.10. Adjusted R 2 values with P value, unstandardized coefficient (β), and unique partial correlation coefficients were used to estimate the contribution of each metrics to the models. A two-sided P < 0.05 was set as statistical significance.

Demographics and Clinical
Characteristics. Demographics and clinical characteristics of the participants are presented in Table 1. irty-seven individuals (28 male, aged 49.78 ± 10.26 years) with subacute stroke were recruited in this study from December 2019 to January 2021 ( Figure 1

Correlations between Clinical and Kinematic Measures.
Correlations between clinical and kinematic measures are shown in Table 2. Mean velocity strongly correlated with the FMA-UE (r � 0.85, P < 0.01) and ARAT (r � 0.80, P < 0.01) and moderately correlated with MBI positively (r � 0.58, P < 0.01). Besides, all the clinical assessments correlated significantly with VP positively (r � 0.55 to 0.81, P < 0.01) and NMU negatively (r � −0.45 to −0.65, P < 0.05). However, MT, TVP%, and NIJ were not significantly associated with the clinical assessments ( Table 2). As shown in Table 2, multicollinearities were observed between MT and NIJ, as well as among VM, VP, and NMU. For that reason, only the VM and NIJ/MT during the FNT task were inputted into the multivariable linear regression models to estimate variation in clinical assessments. e results of multivariable regression analysis of the kinematic metrics against the clinical assessments are presented in Table 3. Backward multiple regression revealed that kinematic variables could explain the largest amount of variance in the assessment of UE motor impairment as measured by FMA-UE. e only significant predictor was the VM, which explained 55% of the FMA-UE variance (F � 20.72, P < 0.01). Moreover, the VM alone showed a significant contribution to the models, accounting for 51% of the ARAT variance (F � 39.10, P < 0.01) and 32% of the MBI variance (F � 8.93, P < 0.01) ( Table 3). Moreover, demographics, including age, gender, type of stroke, and paretic side, showed no significant influence in any regression model.  Journal of Healthcare Engineering    subacute stroke survivors according to the ICF framework. e results indicated that the mean velocity (r � 0.58 to 0.85), peak velocity (r � 0.55 to 0.81), and number of movement units (r � −0.45 to −0.65) were associated with all of the clinical assessments. Mean velocity was entered into the multivariable linear regression models and could estimate 55%, 51%, and 32% of variance in FMA-UE, ARAT, and MBI during the FNT task. Additionally, age, gender, type of stroke, and paretic side had no significant effects on these associations. e previous kinematic literature mainly focused on predicting upper limb motor impairment. Our results extended the value of kinematic scenarios beyond the impairment level according to the ICF framework and suggested that FNT kinematics was more strongly associated with FMA-UE and ARAT than MBI. Lee et al. proposed an automated FMA system and showed high scoring accuracy in 79% of FMA test in nine stroke patients [33]. eir algorithms were shown appropriate for clinical use but lacked clinical interpretability in kinematic results because estimating clinical scale was not the only goal of portable sensors [14,16,34]. In addition, little was known about the associations between clinical activity-related scales and IMU sensors [35]. Our models established predictable correlations between FNT mean velocity via Inertial Measurement Unit and upper extremity motor function after stroke [36].

Discussion
Due to multicollinearity among VM, VP, and NMU, only VM was entered into the multivariable models. Speed variables reflect how efficiently a person controls interaction torques of the agonist/antagonist muscles [32]. Analogous to our results, two studies using robotic device showed significant correlations between movement speed and FMA-UE in individuals with subacute [37] and chronic [38] stroke, respectively. Furthermore, movement smoothness is an important indicator of upper limb motor recovery after stroke [30]. Smoothness parameters evaluate the temporal organization or UE multijoint coordination [39]. In a study early after stroke, smoothness measured by NMU was able to predict upper limb motor recovery over time [40]. According to our results, NIJ was not significantly associated with the clinical assessments. However, smoothness should be interpreted with cautions because a single smoothness parameter may not completely reflect motor recovery of upper extremity [41].
is was analogous with a prior work by Adans-Dester et al., which used eight motor tasks of Wolf Motor Function Test (WMFT) and found satisfactory results to estimate upper limb impairment and activity scales [42]. Although this was in line with our second hypothesis, the difference was small and needed to be further studied [19]. One possible explanation might be that participants had moderate-to-severe upper extremity motor impairment, leading to poor scores on manual dexterity of ARAT items. Future studies should therefore include much kinematic variables and comprehensive tasks at different UE segments to explore the correlations between the assessments. e low variance explained by IMU kinematic variables in MBI could be that the FNT task did not measure distal dexterity of the upper extremity. As a result, variables in the models may not fully capture kinematic information in individuals with stroke [43]. Moreover, MBI is a questionnaire for ADL in a real environment and not an observational measure of UE motor function in an experimental setting. us, participants could use compensatory behaviors or actually the less affected UE to improve MBI scores, which may be difficult to explain with the current kinematic task.
Several limitations of this study should be acknowledged. First, the sample size restricted the number of kinematic variables entered into the multivariable linear regression models.
erefore, future studies could implement other statistical models, such as machine-learning approaches, to investigate the associations between FNT kinematic variables and upper extremity motor function in individuals with stroke [42,44]. Second, this was a cross-sectional study and unable to investigate the longitudinal associations between kinematics and clinical measurements. Finally, the models did not include other kinematic tests and variables concerning trunk and interjoint movements, which may lead to task-related bias and loss of information [45,46].

Conclusions
is study indicates that kinematic variables measured via Inertial Measurement Unit during the finger-to-nose task are associated with upper extremity motor function in individuals with subacute stroke according to the ICF framework. Furthermore, the objective kinematic evaluation may be suitable for predicting clinical measures of motor impairment and capacity to understand upper extremity motor recovery and clinical decision making after stroke.
Data Availability e data files are available from the corresponding author upon reasonable request.

Conflicts of Interest
e authors declare that they have no conflicts of interest.