Exp Brain Res (2015) 233:317–327 DOI 10.1007/s00221-014-4116-5
RESEARCH ARTICLE
Efficiency of visual feedback integration differs between dominant and non-dominant arms during a reaching task Gregory A. Apker · Keith Dyson · Garrett Frantz · Christopher A. Buneo
Received: 12 June 2014 / Accepted: 23 September 2014 / Published online: 10 October 2014 © Springer-Verlag Berlin Heidelberg 2014
Abstract Recent studies have shown that patterns of endpoint variability following double-step reach sequences reflect the influence of both planning and execution-related processes, but are strongly dominated by noise associated with the online updating of movement plans based on visual feedback. However, it is currently unclear whether these results reflect the dominant arm/hemisphere’s postulated specialization for visual feedback processing, or whether these effects reflect a more general “arm/hemisphere independent” preference for visual feedback in the control of reaching. To explore this, twelve subjects performed double-step reach sequences with their dominant and non-dominant arms to targets in 3D space with and without visual feedback of the arm. Variability was quantified using the volumes, aspect ratios, and orientations of 95 % confidence ellipsoids fit to the distributions of reach endpoints. In consonance with previous findings, the availability of visual feedback resulted in ellipsoids that were significantly smaller, had greater aspect ratios, and were more aligned with the depth axis than those performed without visual feedback. Moreover, the effects of vision on aspect ratio and orientation were similar in magnitude for the dominant and non-dominant arms, suggesting that noise associated with planning and execution-related processes is managed in a similar way by the sensorimotor systems of each arm. However, the degree to which vision decreased ellipsoid volume was found to be significantly greater for the dominant arm. This suggests that the feedback control system of the dominant arm uses visual information more efficiently to control reaches to visual targets. G. A. Apker · K. Dyson · G. Frantz · C. A. Buneo (*) School of Biological and Health Systems Engineering, Arizona State University, P.O. Box 879709, Tempe, AZ 85287-9709, USA e-mail: [email protected]
Keywords Vision · Proprioception · Planning · Execution · Variability
Introduction Patterns of movement variability are often used to infer the strategies and mechanisms which underlie sensorimotor integration during goal directed reaching movements. Reaching variability arises in part from noise associated with the encoding and integration of sensory and motor signals (Faisal et al. 2008). Motor-related noise results from both central and peripheral processes and typically manifests in variability directed along the terminal movement vector (van Beers et al. 2004). However, the nature of sensory-related noise is more complex, exhibiting variable properties depending upon the source of the feedback (i.e., visual or proprioceptive). Moreover, in multisensory contexts, integration of feedback sources depends strongly on the relative reliabilities of these signals and stage of motor planning, among other factors (van Beers et al. 1998, 1999; Sober and Sabes 2003). Given this, predicting how sensory and motor processes will interact during movement can be difficult, making interpretation of resulting movement variability problematic (Apker et al. 2010). Investigating the sources of variability and characterizing their interaction is critical for explaining movement errors produced by healthy subjects as well as for understanding the exaggerated variability in neurologically involved patients (Hermsdorfer and Goldenberg 2002; Contreras-Vidal and Buch 2003; Longstaff and Heath 2006; Thies et al. 2009). Studies of arm movement variability typically focus on movements of only one arm, generally the dominant or ‘preferred’ arm. However, it has been proposed that the dominant and non-dominant arms serve distinct functional
13
318
roles, with the dominant arm being used primarily for trajectory control and fine manipulation, and the non-dominant arm being used chiefly for position control and stabilization (Sainburg 2002). Others have suggested that these differences are related to asymmetries in the processing of sensory inputs, with the dominant arm/hemisphere being more efficient at integrating visual feedback and the nondominant arm/hemisphere being more effective at utilizing proprioceptive feedback (Flowers 1975; Roy and Elliott 1986, 1989; Carson et al. 1993; Goble and Brown 2008a). Indeed, in target matching tasks, errors in matching visual targets are generally smaller for the dominant arm, while errors in matching proprioceptive targets are smaller for the non-dominant arm (Goble and Brown 2008b; Goble et al. 2009). Experiments probing more dynamic utilization of proprioception support the idea that the non-dominant arm/ hemisphere is more attuned to the processing of proprioceptive input than the dominant arm/hemisphere (Goble and Brown 2010). Given these observations, it follows that dynamic utilization of visual feedback should differ between arms, an asymmetry that should be most apparent during visually guided reaching tasks. We have recently shown that patterns of endpoint variability following double-step reaching sequences reflect a mixture of sensory and motor processes, but are strongly dominated by noise associated with the online updating of movement plans based on visual feedback (Apker et al. 2010; Apker and Buneo 2012). However, since most subjects used their dominant arms in these studies, it is unclear whether the results reflect the dominant arm/hemisphere’s specialization for processing visual feedback, or whether these effects reflect a more general preference for visual feedback in the control of reaching, i.e., one that is independent of the arm used. To explore this issue, we designed a study to evaluate patterns of arm endpoint variability following visually and non-visually guided reaches executed by the dominant and non-dominant arms. Right- and left-handed subjects performed double-step reaching movements to targets distributed in a 3D virtual workspace. Visual feedback was provided on half of the trials for each arm and was withheld for the remaining half. The size, shape, and orientation of the resulting movement endpoint distributions were used to quantitatively characterize the patterns of movement variability produced by each arm. We anticipated that the nature of any differences between the arms would differ somewhat with respect to what has been found in position-matching studies, given the additional contribution of motor-related feedback and uncertainty associated with our task (van Beers et al. 2004; Apker et al. 2010; Apker and Buneo 2012). However, we hypothesized that visual feedback would have a greater effect on movement variability of the dominant arm than of the non-dominant arm, consistent
13
Exp Brain Res (2015) 233:317–327
with observed asymmetries between arms during matching tasks (Goble and Brown 2008b, 2010; Goble et al. 2009).
Methods Subjects Twelve (12) subjects (5 women, 7 men) between the ages of eighteen and twenty-seven were recruited to perform the experiment. The experiment complied with and was approved by the Arizona State University Institutional Review Board (IRB) before subjects were recruited and data were collected. All subjects read and signed an IRB approved consent form prior to participating. Subjects were briefed on the experimental procedures and expectations for moving within the virtual environment but were naïve to the purpose of the study. Prior to beginning the reaching tasks, subjects were asked to complete the revised Edinburgh Handedness Inventory (Oldfield 1971), which was scored to determine not only the identity of the dominant arm, but also the extent of arm dominance. Apparatus The experimental apparatus used in this experiment is shown in Fig. 1. In this setup, a 3D monitor (Dimension Technologies, Rochester, NY) projected images onto a mirror embedded in a large metal shield. The monitor and mirror were used to display a virtual environment to the subject, and the metal shield housing the mirror also served to block the subject’s hand from view while moving in the workspace. Seat height was adjusted so that subjects could comfortably use a chinrest, which ensured proper orientation of the head with respect to the mirror throughout the experiment. Movements of the fingertip were tracked using an active-LED-based motion tracking system (Visualeyez™ VZ-3000 motion tracker (Phoenix Technologies Inc, Burnaby, British Columbia; 150 Hz sampling rate; 0.5-mm spatial resolution). Fingertip position was fed back to the subject in near-real time using a virtual reality (VR) environment developed in Vizard® (WorldViz LLC. Santa Barbara, CA). A single LED placed on the subject’s fingertip during the experiment relayed position data to the control computer. The VR environment consisted of green spheres ~5 cm in diameter (when in the vertical plane of the starting position) to represent the fingertip, starting, and target positions and a wireframe cube that enclosed the workspace, which aided depth perception. The center point of the cube was located at the starting position, and the side lengths of the cube were great enough that all targets and motions were contained within the cube.
Exp Brain Res (2015) 233:317–327
(a) Chin rest
3D Monitor
319
(b)
Frontal
Display mirror T2ccw
T1left z
T1up
S
Sagittal
T2in
T2cw
T1right
T1up
T2out
S
y x
T2
y
Workspace
x
T1down
T2
T2
y
T1down
-z
Frontal
(c)
T2
Horizontal
T1up T2
T1left
T2
y x
S
T1down
T2ccw
T2
T1right
T1left
T2cw
T2
z
T2in
S
T1right T2out
x
Fig. 1 Experimental apparatus and example movement sequences (Apker and Buneo 2012). a Experimental apparatus. b Frontal and sagittal plane views of the 4 potential movement sequences associated with T1up. The second movement in the sequence is directed to
1 of 4 secondary targets located clockwise (T2cw), counterclockwise (T2ccw), inward (T2in), or outward (T2out) from T1. c Frontal and horizontal plane views of the 4 potential movement sequences associated with T1right
Experimental design
start of a trial. Display of the starting position at the beginning of each trial cued the subjects to move their fingertip to the starting position. Visual feedback was always present during this segment of the trial, so subjects were able to visually align with the starting position. After staying within a 4 cm window centered on the starting position for 350 ms, a pseudo-randomly selected T1 would appear, cueing the first motion. Immediately upon leaving the starting position window, the T1 would disappear and a corresponding T2 would appear, cueing the second motion. Since the target change occurred at very near peak hand velocity toward T1, the potential influence of saccadic suppression was largely attenuated. During V trials, visual feedback of fingertip position was relayed throughout the trial. During NV trials, as soon as T1 appeared, visual feedback of fingertip was withdrawn for the remainder of the trial. Subjects were instructed to move quickly and accurately to the target and to avoid adjusting fingertip position during the holding period. A trial was considered successful after the subject moved to the appropriate T2 and remained within a 5 cm window around that position for 350 ms. Subjects were informed of successful trial completion by an auditory tone. The subject was asked to successfully complete each combination of variables fifteen (15) times for each arm. Only data from successful trials were retained for data analysis.
Subjects executed a two-step movement sequence to visual targets in 3D space. Sequences entailed a movement to one of four primary targets (T1) centered around the starting position in the frontal plane, followed by a final movement to one of the four secondary targets (T2) evenly spaced around each T1. As illustrated in Fig. 1, the arrangement of the T2s was such that the second movement in a sequence was either within the same frontal plane as the first movement (either clockwise or counterclockwise w.r.t. a T1) or involved movements largely along the depth axis (inward or outward w.r.t. a T1); these two sequences are referred to as ‘frontal sequences’ or ‘depth sequences,’ respectively. All targets were arranged on the surface of a 9 cm diameter invisible sphere that was centered on the starting position, ensuring that each movement sequence entailed the same required movement distance. Sequences were performed under one of two feedback conditions: with visual feedback of fingertip position provided throughout movement (visual (V) condition) or without visual feedback of the fingertip during movement (non-visual (NV) condition). In both conditions, once the secondary target appeared, it remained illuminated until the end of the trial. Task parameters T1, T2, and visual condition (V or NV) were pseudo-randomly selected for each trial and subjects had no prior knowledge of these parameters prior to the
13
320
Exp Brain Res (2015) 233:317–327
Data analysis Movement data were sorted according to subject, feedback condition, target sequence (i.e., combination of T1 and T2), and arm used and then smoothed offline using a digital low-pass filter (5-point moving average). Movement endpoints were identified as the point at which the tangential movement velocity (calculated by differentiating the position data along the movement path) fell below 5 % of its peak value for movements to T2. As in our previous experiments, constant errors were not explored in detail here. Analysis focused instead on the variable errors, which are thought to provide information specific to planning and execution-related phenomena (Gordon et al. 1994; McIntyre et al. 1998; Carrozzo et al. 1999; van Beers et al. 2004). Variable errors associated with a given axis and T2 position (σt) were calculated as follows: 1 nt i ¯ 2 ht − ht σt = (1) i=1 nt where h¯ t represents the mean endpoint position for a given T2 position t, hti represents the corresponding endpoint position on trial i, and nt represents the number of trials. Principal components analysis (PCA) was used to quantify several spatial characteristics of the endpoint distributions. To do this, the 95 % tolerance ellipsoids associated with each endpoint distribution were first computed as follows (Morrison 1990; McIntyre et al. 1998):
T95%
(n + 1)(n − k) F0.05,q,n−q−k+1, H =q n(n − q − k + 1)
(2)
where the dimensionality q = 3, the number of target positions k = 1, and H is the covariance matrix. PCA of the resulting T matrix produced the eigenvalues and eigenvectors which were used to quantify the sizes, shapes, and orientations of the endpoint distributions (see below). Ninety-five percent confidence ellipses and ellipsoids were calculated for visualization of performance using MATLAB code based on the Khachiyan algorithm (Khachiyan and Todd 1993; Khachiyan 1996), as implemented by Nima Moshtagh. Endpoint distributions for each arm and sequence were compared based on the size, shape, and orientation of the corresponding tolerance ellipsoids. The size of each ellipsoid was quantified by its volume (V):
V=
4π abc 3
(3)
where a represents the radius of the major axis of the 95 % confidence ellipsoid, and b and c represent the radii of the minor axes. The aspect ratio (shape) of each ellipsoid was defined as the ratio of the radius of the major axis of the ellipsoid to the sum of the radii of the minor axes. Ellipsoid
13
orientation was defined by the orientation of the primary axis of variability, given by the orientation of the corresponding first principle component derived from PCA. Relative contribution of vision In order to quantify the effect of vision on ellipsoid volumes and aspect ratios, a Relative Contribution of Vision (RCV) index was calculated for each parameter as follows:
RCVVolume,h,i =
VolumeNV,i VolumeV,i
RCVAspect Ratio,h,i =
Aspect RatioV,i Aspect RatioNV,i
where h represents the hand/arm and i represents the T2 sequence. For ellipsoid volume, RCV values greater than 1 indicate the factor by which total variability increases in the absence of visual feedback. For aspect ratio, the numerator and denominator are switched so that RCV values greater than 1 indicate the factor by which endpoint distributions narrowed in the presence of visual feedback. Thus, for both aspect ratio and volume, larger RCV values indicate a greater influence of vision. Statistical analyses Differences in the size and shape of the endpoint ellipsoids between arms and within a particular visual condition were assessed using the Mann–Whitney U test. Differences in ellipsoid size and shape between visual conditions and within each arm were assessed using a Wilcoxon signedrank test on the volumes and aspect ratios calculated in the V and NV conditions. Effects of vision on orientation were determined using a one-sample nonparametric circular median test on the distribution of differences in ellipsoid orientation between conditions, the latter obtained from the dot product of the first eigenvectors derived from the PCA. To assess whether the effects of vision on ellipsoid size and shape differed between arms, a Wilcoxon signed-rank test was performed on the RCVVolume and RCVAspect Ratio indices calculated for each arm. To determine whether the effects of vision on ellipsoid orientation differed between arms, a two-sample nonparametric circular median test was performed on the distributions of differences in orientation (between V and NV conditions) that were obtained for each arm. We also investigated the relationship between RCV indexes and EHI scores in this study. Following the classification scheme of Knecht and colleagues (Knecht et al. 2000), subjects were assigned to one of two groups: strongly dominant (unsigned EHI score of 75
Exp Brain Res (2015) 233:317–327 Dominant
T1UP
Non-Dominant Vision
10
10
Non-Vision
5
5
Depth
Depth
Fig. 2 Average trajectories and endpoint ellipsoids for selected sequences in the V (black) and NV (gray) conditions performed by a single representative subject. a D arm. b ND arm. For both arms, variable errors are smaller and more elongated in depth in the V condition than those in the NV condition
321
S
0
-5
-5
-10
-10 -10
-5
0
5
S
0
10
-10
-5
Horizontal
10
10
5
5
Vertical
Vertical
T1RIGHT
S
0
0
5
10
Horizontal
0
S
-5
-5 -10
-10 -10
-5
0
5
10
-10
Depth
or greater; N = 7) or weakly dominant (unsigned EHI score less than 75; N = 5). Subjects corresponding RCV Volume and RCVAspect Ratio indices were then compared statistically using the Mann–Whitney U test. The significance level for all statistical tests used in this study was α = 0.05.
Results Handedness Edinburgh Handedness Inventory (EHI) values (Oldfield 1971), which range from −100 for strongly left-handed to +100 for strongly right-handed subjects, were used to determine the degree of dominance of their right or left hand for each subject. For the twelve subjects included in the present study, scores ranged from −82.3 to +100, indicating a broad range in the level of handedness in our population. Notably, the seven right-handed subjects had a significant tendency for stronger dominance (93.75 ± 6.2, mean and standard error, respectively) compared to lefthanded individuals (−60 ± 7.3). This reflects the classic
-5
0
5
10
Depth
distribution of EHI scores for right- and left-handed individuals (Oldfield 1971). Although the main focus of this study was investigating differences in the utilization of vision between the D and ND arms, we also compared endpoint errors between arms within each visual condition. We found only inconsistent differences between the arms in the two conditions with the only consistent, significant difference being between ellipsoid volumes of the D and ND arms in the visual condition for frontal sequences, with the ellipsoid volumes of the D arm being slightly smaller (Mann–Whitney U test, z-value = 2.68, p
RESEARCH ARTICLE
Efficiency of visual feedback integration differs between dominant and non-dominant arms during a reaching task Gregory A. Apker · Keith Dyson · Garrett Frantz · Christopher A. Buneo
Received: 12 June 2014 / Accepted: 23 September 2014 / Published online: 10 October 2014 © Springer-Verlag Berlin Heidelberg 2014
Abstract Recent studies have shown that patterns of endpoint variability following double-step reach sequences reflect the influence of both planning and execution-related processes, but are strongly dominated by noise associated with the online updating of movement plans based on visual feedback. However, it is currently unclear whether these results reflect the dominant arm/hemisphere’s postulated specialization for visual feedback processing, or whether these effects reflect a more general “arm/hemisphere independent” preference for visual feedback in the control of reaching. To explore this, twelve subjects performed double-step reach sequences with their dominant and non-dominant arms to targets in 3D space with and without visual feedback of the arm. Variability was quantified using the volumes, aspect ratios, and orientations of 95 % confidence ellipsoids fit to the distributions of reach endpoints. In consonance with previous findings, the availability of visual feedback resulted in ellipsoids that were significantly smaller, had greater aspect ratios, and were more aligned with the depth axis than those performed without visual feedback. Moreover, the effects of vision on aspect ratio and orientation were similar in magnitude for the dominant and non-dominant arms, suggesting that noise associated with planning and execution-related processes is managed in a similar way by the sensorimotor systems of each arm. However, the degree to which vision decreased ellipsoid volume was found to be significantly greater for the dominant arm. This suggests that the feedback control system of the dominant arm uses visual information more efficiently to control reaches to visual targets. G. A. Apker · K. Dyson · G. Frantz · C. A. Buneo (*) School of Biological and Health Systems Engineering, Arizona State University, P.O. Box 879709, Tempe, AZ 85287-9709, USA e-mail: [email protected]
Keywords Vision · Proprioception · Planning · Execution · Variability
Introduction Patterns of movement variability are often used to infer the strategies and mechanisms which underlie sensorimotor integration during goal directed reaching movements. Reaching variability arises in part from noise associated with the encoding and integration of sensory and motor signals (Faisal et al. 2008). Motor-related noise results from both central and peripheral processes and typically manifests in variability directed along the terminal movement vector (van Beers et al. 2004). However, the nature of sensory-related noise is more complex, exhibiting variable properties depending upon the source of the feedback (i.e., visual or proprioceptive). Moreover, in multisensory contexts, integration of feedback sources depends strongly on the relative reliabilities of these signals and stage of motor planning, among other factors (van Beers et al. 1998, 1999; Sober and Sabes 2003). Given this, predicting how sensory and motor processes will interact during movement can be difficult, making interpretation of resulting movement variability problematic (Apker et al. 2010). Investigating the sources of variability and characterizing their interaction is critical for explaining movement errors produced by healthy subjects as well as for understanding the exaggerated variability in neurologically involved patients (Hermsdorfer and Goldenberg 2002; Contreras-Vidal and Buch 2003; Longstaff and Heath 2006; Thies et al. 2009). Studies of arm movement variability typically focus on movements of only one arm, generally the dominant or ‘preferred’ arm. However, it has been proposed that the dominant and non-dominant arms serve distinct functional
13
318
roles, with the dominant arm being used primarily for trajectory control and fine manipulation, and the non-dominant arm being used chiefly for position control and stabilization (Sainburg 2002). Others have suggested that these differences are related to asymmetries in the processing of sensory inputs, with the dominant arm/hemisphere being more efficient at integrating visual feedback and the nondominant arm/hemisphere being more effective at utilizing proprioceptive feedback (Flowers 1975; Roy and Elliott 1986, 1989; Carson et al. 1993; Goble and Brown 2008a). Indeed, in target matching tasks, errors in matching visual targets are generally smaller for the dominant arm, while errors in matching proprioceptive targets are smaller for the non-dominant arm (Goble and Brown 2008b; Goble et al. 2009). Experiments probing more dynamic utilization of proprioception support the idea that the non-dominant arm/ hemisphere is more attuned to the processing of proprioceptive input than the dominant arm/hemisphere (Goble and Brown 2010). Given these observations, it follows that dynamic utilization of visual feedback should differ between arms, an asymmetry that should be most apparent during visually guided reaching tasks. We have recently shown that patterns of endpoint variability following double-step reaching sequences reflect a mixture of sensory and motor processes, but are strongly dominated by noise associated with the online updating of movement plans based on visual feedback (Apker et al. 2010; Apker and Buneo 2012). However, since most subjects used their dominant arms in these studies, it is unclear whether the results reflect the dominant arm/hemisphere’s specialization for processing visual feedback, or whether these effects reflect a more general preference for visual feedback in the control of reaching, i.e., one that is independent of the arm used. To explore this issue, we designed a study to evaluate patterns of arm endpoint variability following visually and non-visually guided reaches executed by the dominant and non-dominant arms. Right- and left-handed subjects performed double-step reaching movements to targets distributed in a 3D virtual workspace. Visual feedback was provided on half of the trials for each arm and was withheld for the remaining half. The size, shape, and orientation of the resulting movement endpoint distributions were used to quantitatively characterize the patterns of movement variability produced by each arm. We anticipated that the nature of any differences between the arms would differ somewhat with respect to what has been found in position-matching studies, given the additional contribution of motor-related feedback and uncertainty associated with our task (van Beers et al. 2004; Apker et al. 2010; Apker and Buneo 2012). However, we hypothesized that visual feedback would have a greater effect on movement variability of the dominant arm than of the non-dominant arm, consistent
13
Exp Brain Res (2015) 233:317–327
with observed asymmetries between arms during matching tasks (Goble and Brown 2008b, 2010; Goble et al. 2009).
Methods Subjects Twelve (12) subjects (5 women, 7 men) between the ages of eighteen and twenty-seven were recruited to perform the experiment. The experiment complied with and was approved by the Arizona State University Institutional Review Board (IRB) before subjects were recruited and data were collected. All subjects read and signed an IRB approved consent form prior to participating. Subjects were briefed on the experimental procedures and expectations for moving within the virtual environment but were naïve to the purpose of the study. Prior to beginning the reaching tasks, subjects were asked to complete the revised Edinburgh Handedness Inventory (Oldfield 1971), which was scored to determine not only the identity of the dominant arm, but also the extent of arm dominance. Apparatus The experimental apparatus used in this experiment is shown in Fig. 1. In this setup, a 3D monitor (Dimension Technologies, Rochester, NY) projected images onto a mirror embedded in a large metal shield. The monitor and mirror were used to display a virtual environment to the subject, and the metal shield housing the mirror also served to block the subject’s hand from view while moving in the workspace. Seat height was adjusted so that subjects could comfortably use a chinrest, which ensured proper orientation of the head with respect to the mirror throughout the experiment. Movements of the fingertip were tracked using an active-LED-based motion tracking system (Visualeyez™ VZ-3000 motion tracker (Phoenix Technologies Inc, Burnaby, British Columbia; 150 Hz sampling rate; 0.5-mm spatial resolution). Fingertip position was fed back to the subject in near-real time using a virtual reality (VR) environment developed in Vizard® (WorldViz LLC. Santa Barbara, CA). A single LED placed on the subject’s fingertip during the experiment relayed position data to the control computer. The VR environment consisted of green spheres ~5 cm in diameter (when in the vertical plane of the starting position) to represent the fingertip, starting, and target positions and a wireframe cube that enclosed the workspace, which aided depth perception. The center point of the cube was located at the starting position, and the side lengths of the cube were great enough that all targets and motions were contained within the cube.
Exp Brain Res (2015) 233:317–327
(a) Chin rest
3D Monitor
319
(b)
Frontal
Display mirror T2ccw
T1left z
T1up
S
Sagittal
T2in
T2cw
T1right
T1up
T2out
S
y x
T2
y
Workspace
x
T1down
T2
T2
y
T1down
-z
Frontal
(c)
T2
Horizontal
T1up T2
T1left
T2
y x
S
T1down
T2ccw
T2
T1right
T1left
T2cw
T2
z
T2in
S
T1right T2out
x
Fig. 1 Experimental apparatus and example movement sequences (Apker and Buneo 2012). a Experimental apparatus. b Frontal and sagittal plane views of the 4 potential movement sequences associated with T1up. The second movement in the sequence is directed to
1 of 4 secondary targets located clockwise (T2cw), counterclockwise (T2ccw), inward (T2in), or outward (T2out) from T1. c Frontal and horizontal plane views of the 4 potential movement sequences associated with T1right
Experimental design
start of a trial. Display of the starting position at the beginning of each trial cued the subjects to move their fingertip to the starting position. Visual feedback was always present during this segment of the trial, so subjects were able to visually align with the starting position. After staying within a 4 cm window centered on the starting position for 350 ms, a pseudo-randomly selected T1 would appear, cueing the first motion. Immediately upon leaving the starting position window, the T1 would disappear and a corresponding T2 would appear, cueing the second motion. Since the target change occurred at very near peak hand velocity toward T1, the potential influence of saccadic suppression was largely attenuated. During V trials, visual feedback of fingertip position was relayed throughout the trial. During NV trials, as soon as T1 appeared, visual feedback of fingertip was withdrawn for the remainder of the trial. Subjects were instructed to move quickly and accurately to the target and to avoid adjusting fingertip position during the holding period. A trial was considered successful after the subject moved to the appropriate T2 and remained within a 5 cm window around that position for 350 ms. Subjects were informed of successful trial completion by an auditory tone. The subject was asked to successfully complete each combination of variables fifteen (15) times for each arm. Only data from successful trials were retained for data analysis.
Subjects executed a two-step movement sequence to visual targets in 3D space. Sequences entailed a movement to one of four primary targets (T1) centered around the starting position in the frontal plane, followed by a final movement to one of the four secondary targets (T2) evenly spaced around each T1. As illustrated in Fig. 1, the arrangement of the T2s was such that the second movement in a sequence was either within the same frontal plane as the first movement (either clockwise or counterclockwise w.r.t. a T1) or involved movements largely along the depth axis (inward or outward w.r.t. a T1); these two sequences are referred to as ‘frontal sequences’ or ‘depth sequences,’ respectively. All targets were arranged on the surface of a 9 cm diameter invisible sphere that was centered on the starting position, ensuring that each movement sequence entailed the same required movement distance. Sequences were performed under one of two feedback conditions: with visual feedback of fingertip position provided throughout movement (visual (V) condition) or without visual feedback of the fingertip during movement (non-visual (NV) condition). In both conditions, once the secondary target appeared, it remained illuminated until the end of the trial. Task parameters T1, T2, and visual condition (V or NV) were pseudo-randomly selected for each trial and subjects had no prior knowledge of these parameters prior to the
13
320
Exp Brain Res (2015) 233:317–327
Data analysis Movement data were sorted according to subject, feedback condition, target sequence (i.e., combination of T1 and T2), and arm used and then smoothed offline using a digital low-pass filter (5-point moving average). Movement endpoints were identified as the point at which the tangential movement velocity (calculated by differentiating the position data along the movement path) fell below 5 % of its peak value for movements to T2. As in our previous experiments, constant errors were not explored in detail here. Analysis focused instead on the variable errors, which are thought to provide information specific to planning and execution-related phenomena (Gordon et al. 1994; McIntyre et al. 1998; Carrozzo et al. 1999; van Beers et al. 2004). Variable errors associated with a given axis and T2 position (σt) were calculated as follows: 1 nt i ¯ 2 ht − ht σt = (1) i=1 nt where h¯ t represents the mean endpoint position for a given T2 position t, hti represents the corresponding endpoint position on trial i, and nt represents the number of trials. Principal components analysis (PCA) was used to quantify several spatial characteristics of the endpoint distributions. To do this, the 95 % tolerance ellipsoids associated with each endpoint distribution were first computed as follows (Morrison 1990; McIntyre et al. 1998):
T95%
(n + 1)(n − k) F0.05,q,n−q−k+1, H =q n(n − q − k + 1)
(2)
where the dimensionality q = 3, the number of target positions k = 1, and H is the covariance matrix. PCA of the resulting T matrix produced the eigenvalues and eigenvectors which were used to quantify the sizes, shapes, and orientations of the endpoint distributions (see below). Ninety-five percent confidence ellipses and ellipsoids were calculated for visualization of performance using MATLAB code based on the Khachiyan algorithm (Khachiyan and Todd 1993; Khachiyan 1996), as implemented by Nima Moshtagh. Endpoint distributions for each arm and sequence were compared based on the size, shape, and orientation of the corresponding tolerance ellipsoids. The size of each ellipsoid was quantified by its volume (V):
V=
4π abc 3
(3)
where a represents the radius of the major axis of the 95 % confidence ellipsoid, and b and c represent the radii of the minor axes. The aspect ratio (shape) of each ellipsoid was defined as the ratio of the radius of the major axis of the ellipsoid to the sum of the radii of the minor axes. Ellipsoid
13
orientation was defined by the orientation of the primary axis of variability, given by the orientation of the corresponding first principle component derived from PCA. Relative contribution of vision In order to quantify the effect of vision on ellipsoid volumes and aspect ratios, a Relative Contribution of Vision (RCV) index was calculated for each parameter as follows:
RCVVolume,h,i =
VolumeNV,i VolumeV,i
RCVAspect Ratio,h,i =
Aspect RatioV,i Aspect RatioNV,i
where h represents the hand/arm and i represents the T2 sequence. For ellipsoid volume, RCV values greater than 1 indicate the factor by which total variability increases in the absence of visual feedback. For aspect ratio, the numerator and denominator are switched so that RCV values greater than 1 indicate the factor by which endpoint distributions narrowed in the presence of visual feedback. Thus, for both aspect ratio and volume, larger RCV values indicate a greater influence of vision. Statistical analyses Differences in the size and shape of the endpoint ellipsoids between arms and within a particular visual condition were assessed using the Mann–Whitney U test. Differences in ellipsoid size and shape between visual conditions and within each arm were assessed using a Wilcoxon signedrank test on the volumes and aspect ratios calculated in the V and NV conditions. Effects of vision on orientation were determined using a one-sample nonparametric circular median test on the distribution of differences in ellipsoid orientation between conditions, the latter obtained from the dot product of the first eigenvectors derived from the PCA. To assess whether the effects of vision on ellipsoid size and shape differed between arms, a Wilcoxon signed-rank test was performed on the RCVVolume and RCVAspect Ratio indices calculated for each arm. To determine whether the effects of vision on ellipsoid orientation differed between arms, a two-sample nonparametric circular median test was performed on the distributions of differences in orientation (between V and NV conditions) that were obtained for each arm. We also investigated the relationship between RCV indexes and EHI scores in this study. Following the classification scheme of Knecht and colleagues (Knecht et al. 2000), subjects were assigned to one of two groups: strongly dominant (unsigned EHI score of 75
Exp Brain Res (2015) 233:317–327 Dominant
T1UP
Non-Dominant Vision
10
10
Non-Vision
5
5
Depth
Depth
Fig. 2 Average trajectories and endpoint ellipsoids for selected sequences in the V (black) and NV (gray) conditions performed by a single representative subject. a D arm. b ND arm. For both arms, variable errors are smaller and more elongated in depth in the V condition than those in the NV condition
321
S
0
-5
-5
-10
-10 -10
-5
0
5
S
0
10
-10
-5
Horizontal
10
10
5
5
Vertical
Vertical
T1RIGHT
S
0
0
5
10
Horizontal
0
S
-5
-5 -10
-10 -10
-5
0
5
10
-10
Depth
or greater; N = 7) or weakly dominant (unsigned EHI score less than 75; N = 5). Subjects corresponding RCV Volume and RCVAspect Ratio indices were then compared statistically using the Mann–Whitney U test. The significance level for all statistical tests used in this study was α = 0.05.
Results Handedness Edinburgh Handedness Inventory (EHI) values (Oldfield 1971), which range from −100 for strongly left-handed to +100 for strongly right-handed subjects, were used to determine the degree of dominance of their right or left hand for each subject. For the twelve subjects included in the present study, scores ranged from −82.3 to +100, indicating a broad range in the level of handedness in our population. Notably, the seven right-handed subjects had a significant tendency for stronger dominance (93.75 ± 6.2, mean and standard error, respectively) compared to lefthanded individuals (−60 ± 7.3). This reflects the classic
-5
0
5
10
Depth
distribution of EHI scores for right- and left-handed individuals (Oldfield 1971). Although the main focus of this study was investigating differences in the utilization of vision between the D and ND arms, we also compared endpoint errors between arms within each visual condition. We found only inconsistent differences between the arms in the two conditions with the only consistent, significant difference being between ellipsoid volumes of the D and ND arms in the visual condition for frontal sequences, with the ellipsoid volumes of the D arm being slightly smaller (Mann–Whitney U test, z-value = 2.68, p
