Abstract— Deficits in upper limb function are common among patients with Traumatic Brain Injury (TBI). Accordingly, new technologies, such as virtual reality (VR), are being developed to further upper limb rehabilitation. The study described here successfully trialed a table-top VR-based system (called Elements). Two patients with TBI participated in case-studies using a multiple-baseline, AB time-sequence design; the intervention consisted of 12 1-hour sessions. Performance was measured on both system-rated measures and standardized tests of functional skill. Time-sequence plots for each patient were first sight inspected for trends; this was followed by split-middle trend analysis. Participants demonstrated significant improvements in their movement accuracy, efficiency, and bimanual dexterity; and mixed improvement on speed and other measures of movement skill. Taken together, these findings demonstrate that the Elements system facilitated motor learning in both TBI patients. Larger scale clinical trials are now deemed a viable step in further validating the system.

I. INTRODUCTION A. Overview

The application of virtual reality (VR) to assessment and rehabilitation has gained considerable impetus in recent years [1]. In an earlier paper, we described the development of a low-cost system (called Elements) for use with Traumatic Brain Injury (TBI). Elements uses low-end technologies to achieve stable movement tracking, flexible presentation of virtual environments (VEs) and augmented feedback (AF), and automated recording of performance data. This system was designed to augment current approaches to rehab. This paper presents a clinical evaluation of this system using a case study design with multiple baselines.

B. Traumatic Brain Injury

Manuscript received April 16, 2008. This work was supported in part by an

Australian Research Council (ARC) Linkage Grant LP0562622, and Synapse Grant awarded by the Australia Council for the Arts.

N. Mumford is a PhD candidate in Psychology, RMIT University. J. Duckworth is an artist and PhD candidate with the School of Creative

Media, RMIT University, Melbourne Australia. R. Eldridge and H. Rudolph are with the School of Electrical and

Computer Engineering, RMIT University, Melbourne Australia. M. Gugliemetti is an artist and Lecturer with the School of Art and Design,

Monash University. P. Thomas is with the School of Curriculum, Teaching and Learning,

Griffith University, Brisbane, Australia (email: p.thomas@griffith.edu.au). D. Shum is with the School of Psychology, Griffith University, Brisbane, Australia (email: d.shum@griffith.edu.au).

P.H. Wilson is with the Division of Psychology, RMIT University, Melbourne, Australia. (Phone 613-9925-2906; fax: 613-9925-3587; email: peter.h.wilson@rmit.edu.au). * Corresponding author.

Patients with TBI commonly experience disability in upper-limb function, including poor timing and accuracy of reaching, reduced ability to grasp and lift objects, or perform fine-motor movements [2]. Impairment to motor planning functions are also common due to damage of brain networks responsible for forming and implementing motor intentions (i.e., premotor cortex, parietal cortex, basal ganglia, and cerebellum) . This is manifest by abnormal kinematics: delayed movement latency, poor trajectory control, and so on [3]. However, the pattern of deficits varies between patients; hence the importance of assessing individual differences and within-person change following treatment. VR may provide an excellent medium for assessing control processes and functional skill, and for designing graded rehabilitation activities. VEs can engage and motivate the TBI patient, automate data collection, and provide greater control over task constraints [4].

The benefits of VR-based training for TBI have been widely claimed, although most supporting data is drawn from stroke patients (see [5] for review). To date, VR-rehab studies among TBI patients have generally been exploratory. Several researchers have verified the usability of VR programs designed to assess cognitive function [6, 7], and to re-teach functional skills such as driving and cooking [8-10]. Only two studies have assessed the clinical benefits of VR-augmented rehab among TBI patients: one group successfully improved participants’ balance [8], the other cognitive function [9], yet we are unaware of any compelling evidence for the effectiveness of VR for the upper limb in TBI.

C. The Elements System The development of the Elements system draws on

emerging trends in Human Computer Interaction (HCI) and the science of movement . As reported elsewhere, we have blended (ecological) approaches to motor control and learning (e.g., Gibson and Turvey) with interaction design to create virtual workspaces for clinical assessment and rehab. The mode of interaction is more direct than traditional VR, and consistent with an embodied view of performance and emerging trends in interactive design (e.g., Krueger ).

Tangible user interfaces (TUIs), for instance, can enable motor learning by reinforcing patients’ sense of position in space via AF [4, 10]. For example, a canonical object (hereafter labeled Object) functions as a TUI; it is manipulated in response to location cues in the VE while being position tracked itself by a high-resolution stereo camera (see [11]).

D. Mixed methodologies for evaluation research

A working assumption of VR-assisted therapy is that it can be used effectively in rehab facilities. If new VR technologies are

A virtual tabletop workspace for upper-limb rehabilitation in Traumatic Brain Injury (TBI): A multiple case study evaluation

Nick Mumford, Jonathan Duckworth, Ross Eldridge, Mark Guglielmetti, Patrick Thomas, David Shum, Heiko Rudolph, Gavin Williams and Peter H. Wilson

978-1-4244-2701-7/08/$25.00 ©2008 IEEE 175

to be implemented, their benefits need to be demonstrated empirically through intervention research. Indeed, the notion of evidence-based practice places an onus upon therapists to use treatments shown to have the greatest potential for recovery.

Case study methods provide a vital research medium for evaluating new and emerging treatment approaches [12]. Classic time-series design involves repeated assessments from a single participant over baseline and treatment phases. These designs permit detailed analysis of individual performance in response to intervention and a flexible means of testing hypotheses (see [13] on the merits of case-study research). In short, time series designs offer flexibility of administration, and are clinically relevant by focusing on within-patient change.

The aim of the present study was to evaluate a new VR-augmented rehab system designed to improve upper limb function among TBI patients by conducting a series of single participant case-studies, using an AB time-sequence method. Specifically, we predicted that following a 12-session course of VR-augmented therapy, concurrent with normal rehabilitation, TBI participants’ upper limb function would improve compared with baseline performance.

II. METHOD A. Participants

Two TBI patients were recruited from the Epworth Hospital in Melbourne, Australia. Patients were referred by the Senior Physiotherapist. Inclusion criteria were as follows: age under 50 years; a score of at least two for muscle activity as measured on the Oxford scale; and a cognitive capacity to provide informed consent (as decided by physiotherapy and occupational therapy staff) and to understand the rehab program. Details for each participant follow. Participant AN is a male aged 20 years. He experienced TBI as a result of an automobile collision. His TBI occurred 4 months before the study. AN was an inpatient during the study. He presented with ataxia and required assistance to walk. He underwent physical therapy on a daily basis focusing mainly on mobility and gait; upper-limb training was less intensive. Participant TJ is a male aged 21 years whose TBI also occurred as a result of an automobile collision. He lived with family, then independently as an outpatient while participating in the research. TJ’s TBI occurred 12 months prior to the study. He experienced moderate hemiparesis. TJ was undergoing physical therapy that mainly targeted balance and gait. The study was approved by the Human Research Ethics Committees of Epworth Hospital and RMIT University.

B. Materials 1) Elements system: Hardware: The Elements system runs

on off-the-shelf PC hardware. The PC has an AMD Athlon64 X2 Dual Core 4400+ (2.21GHz) processor and 2GB RAM. The PC is equipped with an nVidia GeForce 7800 graphics card. The VE is displayed onto a 40" LCD panel placed horizontally, with audio cues via stereo speakers; 6mm non-reflective glass covers the panel (see Figure 1 for the system

set up). The movement of the Object (a plastic cylinder, 63mm diameter, 130mm high) is tracked using the Bumblebee2a camera system from PointGrey which has pre-calibrated stereo cameras for accurate depth measurement (see also [1, 11] for the hardware comprising the elements system).

Fig. 1. The Elements system

2) Elements system: Task description and augmented feedback: Task 1 (Bases) consists of the home base and three potential movement targets, 78 mm in diameter. The targets are highlighted in a fixed order (west, north, east). Task 2 (Random Bases) has the same target configuration, but they are highlighted in random order. Task 3 (Chase Task) begins with a blank screen. A target circle then appears randomly in each of nine locations. These locations are configured along three radials emanating from the home base. Task 4 (Go-No Go) uses the same target positions as 3; however, additional distractor targets appear—a hexagon, triangle, and rectangle. Participants are instructed to place the Object on the circle target only.

3) Measured Variables: Three performance variables are recorded automatically by the system. Accuracy of placement is measured as a percentage score, and represents the overlap between the base of the object and the target area; thus 100 percent would be perfect overlap. Movement speed is given by the rate of object movement during task performance, measured in m/s. Movement efficiency is the deviation from the straight-line path between targets. This variable is measured as a percentage score: (total straight-line distance between the sequence of targets ÷ distance moved) x 100. In addition to training, Task 1 and 3 were used as assessment tasks (administered without augmented feedback). The main therapeutic modality used by the Elements system is augmented feedback (AF). AF is the provision of information, over and above the normal flow of visual and movement-related information. In this system, each AF feature is related to one or more of the three movement variables. Thus, the form of AF selected depends on which variables are targeted for improvement (see Table 1 for the AF options).

176

Fig. 2. Screen shot depicting the ripple, trace, and aura AF 4) Standardised Measures of Functional Skill

Upper limb performance was also assessed using two standardized measures: Box and Block Test (BBT). The BBT consists of two connected boxes (54 x 24 cm), separated by a vertical wooden barrier. One box is filled with 150 2.5cm wooden cubes. The goal is to move as many blocks as possible, one at a time, in 60 s using one hand. The BBT has excellent test-retest reliability [14] and predictive validity. McCarron Assessment of Neuromuscular Dysfunction (MAND). Two items assessing bimanual dexterity from the MAND test were used in our study. The bead-threading task requires participants to hold a metal rod in their non-preferred hand and thread as many wooden beads as possible onto it in 30 s. For the nuts-and-bolts task, participants hold a metal nut in their non-preferred hand and screw a bolt into the nut as quickly as possible; a large and small bolt were used. The MAND has demonstrated good reliability and validity. C. Procedure

1) Design of case studies: An AB time-sequence design, with multiple baseline was used. The baseline phase (A) consisted of an initial series of assessments of upper-limb function (VR-system and standardized measures). This phase lasted for four sessions for TJ, and nine for AN. The intervention (B) phase that followed comprised of 12 60-min sessions, conducted over a 4 week period. This structure for treatment is consistent with the American Occupational Therapy Association’s recommendations for treatment of TBI [15]. Performance was assessed at the conclusion of each session. All testing was conducted onsite at the Epworth Hospital. For each intervention session participants performed the tasks using AF; the aim was to improve their level of performance across the three performance variables (accuracy, speed, and efficiency). Participants set general and weekly goals for each variable.

2) Data analysis: The results were presented as time-sequence graphs, and were sight inspected to assess changes between phases. To complement the visual analysis, objective measures of change were employed. Split-middle trend analysis (SMTA) predicts a celeration line from the A phase to the B phase of a time-sequence graph. Cumulative binomial analyses then assess whether a significant number of data points are above or below the celeration line (depending on the index performance) in the B phase [16]. For the binomial tests a probability of 0.5 and significance of α < 0.05 were used. A sample celeration line is presented in figure 6 for AN’s left hand BBT scores.

III. RESULTS

A. System variables 1) Accuracy: The time-sequence plots for accuracy are

presented in Figure 3. Sight inspection of these graphs indicated a trend towards improvement for both participants, using each hand. Follow up SMTA demonstrated significant improvement for TJ (N = 11, p < 0.001 for each hand). Similarly, there was significant improvement in accuracy for AN (N = 10, p = 0.011 for right hand, p < 0.001 for left hand).

2) Speed: Figure 4 presents time-sequence plots for movement speed. Sight inspection indicated improvement on this variable for AN but not for TJ. SMTA confirmed this: AN showed significant improvement for both his left (N = 10, p < 0.001) and right hand (N = 10, p = 0.05), while TJ did not (N = 11, p = 0.999 for left hand, p = 0.994 for the right).

3) Efficiency: Sight inspection of the efficiency data indicated a trend towards improved performance between baseline and intervention phases, for each participant (Figure 5.). Follow up SMTA demonstrated a significant improvement for AN (p < 0.001 for each hand). For TJ, a significant improvement was found for his right hand (p < 0.001), but not left (p = 0.06).

TABLE I DESCRIPTION OF THE ELEMENTS AF FEATRURES, AND THE MOVEMENT

VARIABLES RELATED TO EACH Augmented feedback (AF) Movement variable associated

with AF

Ripple effect for placement When the object is placed on the target, a water ripple animation emanates from that location.

Informs participants' whether the object was placed correctly.

Object trace of trajectory As the object is moved above the screen, a fading trace follows the object’s path on the monitor.

Visual representation of movement efficiency, and accuracy.

Sound pitch for proximity and/or volume for speed As the object approaches the target, a tone increases in magnitude or pitch; pitch can also be varied according to speed. A final ‘click’ sound is emitted when the object is placed on the target.

Reinforces the movement goal, trajectory, and speed, depending on the choice of AF.

Luminescence / Aura effect for proximity to target As the object approaches the correct target, a waxing ‘aura’ appears around the target

Communicates correct movementchoices and accuracy.

177

Fig. 3. Time-sequence plots for movement accuracy on the Elements tasks.

Fig. 4. Time-sequence plots for movement speed.

4) Standardized measures BBT. Sight inspection of the time-sequence plots for BBT performance (Figure 6.) indicated a trend towards improvement for each hand for TJ, and marginal improvement for AN. SMTA showed no significant improvement for TJ with either hand (N = 12, p = 0.98 for left hand and 0.99 for right). For AN, a significant improvement was found for his left hand (p = 0.005), but not right (N = 11, p = 0.997).

Fig. 5. Time-sequence plots for movement efficiency.

Fig. 6. Time-sequence plots for BBT scores. With celeration line for AN’s left hand scores. MAND. The time-sequence plots for the MAND sub-tests showed mixed results (Figures 7 and 8). For the nut and bolt tests, improvement (reduced time) was seen for TJ on both the large and small bolts. Similarly, a trend towards improvement was noted for AN on each version of the test. Follow up SMTA confirmed this: significant improvement was evident for TJ (N = 12, p < 0.001 for each test); for AN, significant improvement was evident for the large bolt (N = 12, p < 0.001), but not the small (N = 12, p = 0.113).

178

Fig. 7. Time-sequence plots for the nuts-and-bolts tests

Fig. 8. Time-sequence plots for the bead threading task For the bead threading task, no notable improvement in performance was visible for either participant. Follow up analyses confirmed this for both TJ (N = 12, p = .997) and AN (N = 11, p = .89).

IV. DISCUSSION The aim of the present study was to assess a new VR-based system for upper-limb rehab, using multiple case studies. In general, the results supported the efficacy of the system at two levels: system-related performance and on standardized

measures of functional skill. These effects were evident despite both patients undergoing physical therapy throughout the study. There was, however, little change on some measures over the course of intervention. Performance trends are discussed in more detail below. A. System-related performance Patient AN showed significant improvement on both left and right hand for all Elements tasks. The results suggest that the AF helped AN improve his performance for both hands; note that his TBI resulted in severe motor impairment to both sides of his body. AN demonstrated significant improvement in both accuracy and speed. During baseline and the start of intervention, AN’s movements of the TUI during transport were fast, but his terminal control was less developed, causing poor accuracy scores (between 50 and 60% during baseline). His accuracy toward the end of training was around 70%, with no decline in speed. This finding indicates genuine improvements in motor control; based on Fitts’ Law [17], we might expect improvement on one of these variables at the expense of the other, but this was not evident. Patient TJ showed less overall improvement on the system variables than AN, but still demonstrated improved motor control. At the beginning of the study it was apparent that TJ’s arm function was generally well rehabilitated. He scored high on the speed and efficiency variables; only accuracy was relatively low at baseline. Therefore, the goal for TJ was to improve his accuracy while maintaining acceptable efficiency and speed. This objective was achieved. TJ showed improved accuracy while maintaining his movement speed (an initial decline in speed was evident early in treatment); this pattern, suggests genuine improvement in motor control, mirrored also by improved movement efficiency. B. Standardised measures of functional skill

Patient AN’s improvement on the large nut-bolt task, but not the small, may indicate a weakness in fine-motor control, since the larger nut/bolt is easier to grasp and manipulate. However, variability on the small version may mask a general training effect. Indeed, greater variability was also noted on the bead-threading task over the intervention phase. One possibility is that fine-motor skill for AN underwent a period of re-organisation due to the intervention. Using an extended program; we would predict a jump in performance as a stable coordination pattern is eventually adopted. This remains to be tested. AN also demonstrated improvement on the BBT only for his non-dominant hand (left). This finding is consistent with a previous study showing that repetition and goal-setting improved arm function in TBI patients’, particularly the non-preferred hand. Patients motivation to improve was seen to explain the effect [18]. This is supported in our study since AN more often chose to perform the elements tasks with his left hand. Patient TJ demonstrated significant improvement on the nuts/bolts tests, but not on the beads task or BBT. For the nuts/bolts tests, TJ improved significantly between phases. However, his performance did appear to level off toward the end of treatment. Observation suggested that TJ did not use

179

information about his performance on this test as motivation, unlike the BBT or threading task; on the latter, he often remarked about how well he performed, suggesting a desire to ‘beat his own score’. Improved performance on the BBT was evident from the start of baseline, and was maintained during intervention. Thus, an extended baseline might be necessary for these measures, or an alternative test be used that shows minimal short-term learning effects. C. General discussion

The improvement demonstrated by both participants demonstrates that providing augmented feedback using a VE can facilitate motor learning. This finding is consistent with the dynamic systems theory of motor control, which states that by presenting graded movement tasks the performers movement ability may be enhanced [17]. Furthermore, the present findings are consistent with studies using similar (but not VR-based), tabletop upper limb training methods with TBI or stroke patients [26-29]. Nevertheless, mixed performance on standardised measures following VR-based upper limb training is consistent with upper limb VR-rehab studies among stroke patients [30-35]. This suggests that further research is required to demonstrate that VR value adds to normal procedures. Accordingly, a large sample randomized clinical trial is planned to follow up this research.

V. CONCLUSION The findings here, although preliminary, suggest that our system can generate functional improvements in the upper-limb function of TBI patients. Improvement on most standardised measures by each participant is an important finding, since clinically meaningful change/learning is only apparent when effects are generalisable. In short, tabletop VR-based systems that incorporate TUIs afford a quite exciting and promising mode of intervention for TBI and related conditions.

REFERENCES

[1] P. H. Wilson, P. Thomas, D. Shum, J. Duckworth, M. Gugliemetti, H. Rudolph, N. Mumford, and R. Eldridge, "A multilevel model for movement rehabilitation in Traumatic Brain Injury (TBI) using Virtual Environments," presented at International Workshop for Virtual Rehabilitation 5, New York, 2006.

[2] P. H. McCrea, J. J. Eng, and A. J. Hodgson, "Biomechanics of reaching: clinical implications for individuals with acquired brain injury," Disability and Rehabilitation, vol. 24, pp. 534 - 541, 2002.

[3] B. A. Wilson, "Traumatic brain injury," in Comprehensive Clinical Psychology, A. S. Bellack and M. Hersen, Eds. Oxford: Elsevier Science Ltd, 1998, pp. 463-486.

[4] A. A. Rizzo, M. T. Schultheis, K. A. Kerns, and C. Mateer, "Analysis of assets for virtual reality applications in neuropsychology," Neuropsychological Rehabilitation, vol. 14, pp. 207-239, 2004.

[5] A. Henderson, N. Korner-Bittensky, and M. Levin, "Virtual reality in stroke rehabilitation: A systematic review of its effectiveness for upper limb motor revovery," Topics in Stroke Rehabilitation, vol. 14, pp. 52-61, 2007.

[6] P. McGeorge, L. Philips, J. Crawford, S. Garden, S. Della-Sala, S. Hamilton, and J. Callender, "using virtual environments in the assessment of executive dysfunction," Presence, vol. 10, pp. 375-383, 2001.

[7] R. J. Matheis, M. T. Schultheis, L. A. Tiersky, J. DeLuca, S. R. Millis, and A. Rizzo, "Is learning and memory different in a virtyal environment?," The Clinical Neuropsychologist, vol. 21, pp. 146–161, 2007.

[8] H. Sveistrup, J. McComas, M. Thornton, S. Marshall, H. Finestone, A. McCormick, and K. Babulic, "Experimental studies of virtual reality-delivered compared to conventional exercise programs for rehabilitation," Cyberpsychology & Behavior, vol. 6, pp. 245-251, 2003.

[9] M. A. Grealy, D. Johnson, and S. K. Rushton, "Improving cognitive function after brain injury: The use of exercise and virtual reality," Archives of Physical Medicine and Rehabilitation, vol. 80, pp. 661-667, 1999.

[10] M. Holden, "Virtual environments for motor rehabilitation: Review.," Cyberpsychology & Behavior, vol. 8, pp. 187-211, 2005.

[11] P. H. Wilson, J. Duckworth, N. Mumford, R. Eldridge, M. Gugliemetti, P. Thomas, D. Shum, and H. Rudolph, "A virtual tabletop workspace for the assessment of upper limb function in Traumatic Brain Injury (TBI)," presented at Virtual Rehabilitation 2007, Venice, Italy, 2007.

[12] I. Fisher and J. Ziviani, "Explanatory case studies: Implications and applications for clinical research," Australian Occupational Therapy Journal, vol. 51, pp. 185–191, 2004.

[13] T. Kinugasa, E. Cerin, and S. Hooper, "Single-subject research designs and data analyses for assessing elite athletes' conditioning," Sports Medicine, vol. 35, pp. 1035-1050, 2004.

[14] J. Desrosiers, A. Rochette, R. Hebert, and G. Bravo, "The Minnesota Manual Dexterity: Reliability, validity and reference values studies with health elderly people," Canadian Journal of Occupational Therapy, vol. 64, pp. 270-276, 1997.

[15] M. V. Radomski, Occupational therapy practice guidelines for adults with traumatic brain injury. Betheda, MD: The American Occupational Therapy Association Inc, 2000.

[16] M. R. Nourbakhsh and K. J. Ottenbacher, "The statistical analysis of single-subject data: A comparative examination," Physical Therapy, vol. 74, pp. 768-776, 1993.

[17] A. Shumway-Cook and M. H. Woollacott, Motor control: Translating reearch into clinical practice, 3rd ed. New York: Lippincott Williams & Wilkins, 2007.

[18] D. Eastridge and M. P. Mozzoni, "Fluency and functional motor skills following brain injury," Behavioral Interventions, vol. 20, pp. 77-89, 2005.

180