Smoothness discriminates physical from motor imagery practice of arm reaching movements

Physical practice (PP) and motor imagery practice (MP) lead to the execution of fast and accurate arm movements. However, there is currently no information about the influence of MP on movement smoothness, nor about which performance parameters best discriminate these practices. In the current study, we assessed motor performances with an arm pointing task with constrained precision before and after PP (n= 15), MP (n= 15), or no practice (n= 15). We analyzed gains between Pre- and Post-Test for five performance parameters: movement duration, mean and maximal velocities, total displacements, and the number of velocity peaks characterizing movement smoothness. The results showed an improvement of performance after PP and MP for all parameters, except for total displacements. The gains for movement duration, and mean and maximal velocities were statistically higher after PP and MP than after no practice, and comparable between practices. However, motor gains for the number of velocity peaks were higher after PP than MP, suggesting that movements were smoother after PP than after MP. A discriminant analysis also identified the number of velocity peaks as the most relevant parameter that differentiated PP from MP. The current results provide evidence that PP and MP specifically modulate movement smoothness during arm reaching tasks. This difference may rely on online corrections through sensory feedback integration, available during PP but not during MP.


Introduction
Motor skill learning is a central process in everyday life, sustaining adaptation and 2 efficiency of motor behaviors in constantly changing environments. Through physical practice 3 (PP), movements are performed faster, more accurately, and require less energy consumption 4 (Willingham, 1998). Even if continuous and extended practice is known to greatly and durably 5 improve motor performance (Robertson et al., 2004;Kitago and Krakauer, 2013), positive 6 effects of practice can also be observed within a single training session. A growing number of 7 studies indeed observed that a few minutes of practice is sufficient to induce gains in motor 8 performance (e.g., speed and accuracy) on a wide variety of motor tasks, such as sequential 9 finger-tapping or arm-reaching tasks (Karni et al., 1998;Walker et al., 2003;Gentili et al., 2006, 10 2010; Spampinato and Celnik, 2017; Ruffino et al., 2021). This fast learning process, known as 11 motor acquisition, is considered as the first step towards the formation of new and robust motor 12 memories. 13 Although skill learning usually requires PP, alternative forms of practice also exist. 14 Among these, motor imagery, that is the mental simulation of an action without associated 15 motor output, has been largely documented. In fact, mental practice (MP) improves several 16 aspects of motor performance, such as movement accuracy, speed, and force (Yue and Cole,  forward models are neural network that predict the future sensorimotor state (e.g., velocity, 1 movement duration, position) given the current state, the efferent copy of the motor command, 2 and the goal of the movement (Kawato et al., 2003;Wolpert & Flanagan, 2001). Kilteni et al. 3 (2018) strongly supported this assumption by showing that the sensory consequences of 4 imagined movements are predicted during motor imagery. 5 Although PP and MP share common mechanisms, a number of dissimilarities also exist. 6 Perhaps the main difference, at least the most visible, is that during MP there is no sensory 7 feedback about movement (position velocity and acceleration), since the imagined segment in 8 inert. In error-based motor learning process, external sensory feedback is necessary to update 9 the prediction of the internal forward model, via the discrepancy between the predicted state not explain the underline mechanism. Theoretically, it is proposed that the difference between 17 the prediction and the desired outcome based on stored movement representations would be 18 returned as an input to improve the subsequent motor command via a "self-supervised process",  Up to now, performance improvement after MP has been measured, and compared to 22 PP, mainly on three parameters: force, accuracy, and speed. Nonetheless, other parameters are 23 of importance for motor performance, such as movement smoothness that is enhanced after PP 24 (Balasubramanian et al., 2015). Smoothness is related to the form of the velocity profile, which 25 is singled-peaked with one acceleration and one deceleration phase. When the motor command 1 is inaccurate a number of sub-movements are necessary to correct it, creating thus a non-optimal 2 and clumsy movement (Kelso et al., 1979;Ketcham et al., 2002;Ketcham & Stelmach, 2004). 3 Intriguingly, the effects of MP on this parameter are yet unknown. 4 In the current study, we sought to compare PP and MP, considering spatial, temporal, 5 and smoothness parameters. We recorded movement-related trajectories on a graphic tablet 6 from two training groups (PP and MP) and one control group (Ctrl, absence of practice). In line 7 with the literature, we first hypothesized that PP and MP would similarly enhance arm reaching 8 movements, with improvements for all parameters but with greater gains for PP. Alternatively, 9 temporal parameters would similarly improve following PP and MP, as sensory feedback is not 10 a prerequisite in that case, whereas spatial and smoothness parameters would be less improved 11 after MP. Forty-five right-handed adults participated in this study after giving their informed 16 consent. All were healthy and self-reported being free from neurological or physical disorders. 17 The participants were randomly assigned into three groups: the Physical Practice group (PP, n 18 = 15, 6 females, mean age: 25± 2 years old), the Mental Practice group (MP, n = 15, 9 females, 19 mean age: 25 ± 6 years old), and the Control group (Ctrl, n = 15, 7 females, mean age: 28 ± 4 20 years old). 21 Motor imagery vividness of the MP group was assessed by the French version of the 22 revised Movement Imagery Questionnaire 'MIQr' (Lorant and Nicolas, 2004). The MIQr is an 23 8-item self-report questionnaire, in which the participants rate the vividness of their mental 24 6 images using 7-point scales ranging from 1 (really difficult to feel/visualize) to 7 (really easy 1 to feel/visualize), on two modalities (visual imagery and kinesthetic imagery). The average 2 score obtained in the current study was 42.1 ±9.5 (maximum score: 56; minimum score: 8), 3 revealing good imagery capabilities. 4 5 Experimental Device 6 The participants were comfortably seated on a chair in front of a graphic tablet (Intuos4, 7 XL, Wacom, Krefeld, Germany), on which four square-targets (1x1 cm) were presented (see 8 Fig.1A). The distance between the participants' sternum and the first target (T1) was 10 cm. 9 One trial included 10 successive target-to-target movements in the following order: 1 -2 -3 -  Experimental procedure 14 The experimental protocol included two test sessions (PreTest and PostTest) and one 15 training session (Fig. 1C). During the test sessions, all the participants performed 3 actual trials 16 as fast and accurately as possible. During the training session, the participants of the PP were 17 trained as fast and accurately as possible to the task, while those of the MP group were 18 instructed to imagine themselves performing the task as fast and accurately as possible, 19 combining kinesthetic and visual (first-person perspective) imagery modalities. Both training 20 groups performed 60 trials, divided into 6 blocks with 1-min rest between blocks to avoid    Kinematics recording and analysis 9 We recorded movement kinematics at 100Hz using a graphic tablet (Intuos4 XL, 10 Wacom, Krefeld, Germany). The spatial resolution in the present experiment was less than 1 11 mm. Data processing was performed using custom programs written in Matlab (Mathworks, 12 Natick, MA). Position signals in the horizontal plane (X, Y) were low-pass filtered using a 13 digital fifth-order Butterworth filter (zero phase distortion, Matlab 'butter' and 'filtfilt' 14 functions) at a cut-off frequency of 10 Hz. 15 We computed five parameters for each trial: i) movement duration (MD), i.e., the total 1 time elapsed between the moment when the pencil exited the first target and entered the final 2 target; ii) distance, i.e., the total two-dimensional displacement; iii) mean velocity (Vmean), 3 i.e., the average inter-target movement speed; iv) maximal velocity (Vmax), i.e., the average 4 of maximal inter-target movement speed; and v) number of velocity peaks (NbPeaks), i.e., the 5 number of local maxima detected on velocity profiles. We used this parameter to quantify  Electromyographic recording and analysis 17 To verify that muscles were not activated during mental training (MP group), 18 electromyographic (EMG) activity of the biceps brachii (BB) and the triceps brachii (TB) 19 muscles of the right arm were recorded during each imagined trial and compared to EMG 20 activity at rest (10-second recording before training). We used pairs of bipolar silver chloride 21 circular (10-mm diameter) surface electrodes. We positioned the electrodes parallel to muscle 22 fibers, over the middle of the muscles belly with an inter-electrode (center-to-center) distance 1 of 20 mm. The reference electrode was positioned on the medial elbow epicondyle. After 2 shaving and dry-cleaning the skin with alcohol, the impedance was below 5 kΩ. EMG signals 3 were amplified (gain 1000), filtered (with a bandwidth frequency ranging from 10 Hz to 1 kHz), 4 and converted for digital recording and storage with PowerLab 26T and LabChart 7 (AD 5 Instruments). We analyzed the EMG patterns of the muscles by computing their activation level 6 (RMS, root mean square) using the following formula: Statistical analysis 10 We performed the analyses on motor gains to reduce variability between participants, 11 especially at PreTest. We primarily checked the normality of the data (Shapiro-Wilk test), the 12 equality of variance (Levene's test), and the sphericity (Mauchly's test). 13 First, we used unilateral one-sample t-tests compared to the reference value 100 to check 14 whether motor performances improved between Pre and PostTest, for each parameter and each 15 group. Cohen's d was reported for each test and the statistical significance threshold was set at  To compare gains between groups, we then performed one-factor ANOVAs with Group 19 as a between-subject factor and planned comparisons using orthogonal contrasts analysis for 20 each parameter (Howell, 2012). We constructed a contrast matrix to test the following a priori 21 assumptions: i) MP and PP led to better gain when compared to an absence of practice, i.e., Ctrl 22 group (contrast C1), and ii) PP led to better gain when compared to MP (contrast C2). 23 To identify the parameters that best discriminated the groups, we finally realized a  Also, to ensure that participants of the MP group did not activate their muscles during 10 MP, we used Friedman's ANOVAs, comparing the EMG activity of each imagined block with 11 the rest condition, for each muscle (BB and TB). 12 13

14
Summary data 15 Table 1 reports the mean values and the mean gains for the five kinematic parameters.  One sample t-tests 6 Firstly, to check whether motor performances improved between PreTest and PostTest 7 after practices, we used unilateral one-sample t-tests compared to the reference value 100. Table   8 2 reports the results and effect sizes for one-sample t-tests analysis.   One factor ANOVA and contrast analysis 1 Secondly, we compared gains between groups by means of one-factor ANOVAs and planned 2 comparisons. The results are depicted in Figure 2.   Stepwise generalized linear discriminant analysis 10 Finally, we performed a stepwise generalized linear discriminant analysis to identify the 11 parameters that best discriminated the groups. The three groups (PP, MP, and Ctrl) were 12 considered as the dependent variable and the gains for each parameter as the independent 13 variable. The discriminant power of each variable was tested using a forward stepwise 14 approach, revealing that Vmax and NbPeaks significantly contributed to group discrimination  Vmax discriminates practice from the absence of practice, whereas Nbpeaks also discriminates 2 the performance improvement between PP and MP. 3 4 Electromyographic analysis 5 Participants did not activate their muscles during mental training in comparison to rest.  In the current study, we identified the number of velocity peaks, an indicator of 12 movement smoothness, as the most relevant parameter that differentiated PP from MP for an 13 arm pointing task. While classical parameters as movement duration or maximal and mean 14 velocity improved in a comparable extent following both practices, movement smoothness 15 improved following MP but to a lower extent than that after PP. These findings provide relevant 16 information about the specific influence of practice types on motor performance parameters. 17 18 General motor performance improvement 19 Motor performance improvement of arm reaching movement have been widely of temporal parameters, such as movement duration, mean and maximal velocity after both 23 practices, compared to the Control group.

24
The improvement of motor performance following MP could be explained by the ) and a prediction of the related-movement sensory afferents (Grush, 2004). 8 The comparison between these predictions and the stored movement representation would be 9 fed back as input to the controller, leading to an improvement of motor command despite the 10 absence of feedbacks. Interestingly, PP and MP increased mean and maximal velocity to the 11 same extent, but PP greater decreased the number of velocity peaks. These findings provide 12 evidence that PP and MP may improve specifically the parameters of performance for arm 13 reaching tasks. 14 15 Movement smoothness discriminates physical and mental practices 16 The arm reaching movements can be decomposed in two distinct phases: i) an initial 17 impulse phase, involving predictive loops and ii) a final phase, known as the corrective phase, 18 implying online movement corrections (Elliott et al., 2001;Thompson et al., 2007). Kinematic 19 analyses revealed that the first phase can be characterized by one or two high velocity peaks, 20 permitting to quickly get closer to the target, while the second contains low secondary velocity 21 peaks, which are likely to represent corrective sub-movements when approaching the target 22 (Novak et al., 2002). The authors also suggested that PP leads to faster and precise initial 23 movements in order to quickly approach the target and to reduce the number of corrective sub- movements during PP may help to optimize the corrective phase when approaching the target, 10 and therefore to greater reduce NbPeaks in comparison to MP. The absence of sensory 11 feedbacks during imagined movements would be an obstacle to reduce the number of sub-12 movements when actually approaching the target. 13 14

15
In conclusion, the present study provided the first evidence that MP increased 16 smoothness of arm-reaching movement, and that this performance parameter discriminated 17 between PP from MP. Although, no sensory feedbacks are present during imagined movements, 18 the increase of movement velocity would lead to greater smoothness after MP. Further studies 19 could analyse a broader range of movements and tasks (e.g., to perform and/or imagine the 20 movement at different velocities) to better understand the influence of MP on movements 21 parameters.

11
The authors declare that the research was conducted in the absence of any commercial or 12 financial relationship that could be construed as a potential conflict of interest.