Effect Analysis of Virtual-reality Vestibular Rehabilitation based on Eye-tracking

Abstract

Vertigo is one of the most common complaints encountered by physicians and the patients are steadily increasing. These patients are exposed to the risk of secondary accidents such as falls due to vertigo. There are two ways to improve this symptom: medication and rehabilitation. Although temporary symptomatic improvement may be expected in patients treated with medication, vertigo may recur and medication can delay central compensation. In contrast vestibular rehabilitation exploits central mechanisms of neuroplasticity to increase postural stability and enhance visual-vestibular interactions in situations that generate conflicting sensory information. However, vestibular rehabilitation may be compromised by incorrect performance of exercises, and there is a need for active effort and interest from the patient during rehabilitation. To solve these problems, we decided to apply FOVE HMD for eye-tracking and Unity3D to create virtual reality. The proposed eye-tracking based algorithm calculates the concentration of users with eye tracking data and calculates the motion width of the patient with nystagmus, thus the severity of the patient according to the score can be determined. According to our experimental test against healty group and patients group, this result showed the meaningful data to use define the contents result.

1. Introduction

Vertigo is one of the most common complaints encountered by physicians. Balance disorders of vestibular origin have a lifetime prevalence of 7.4% [1], and include benign paroxysmal positional vertigo and Meniere’s disease-associated vertigo. Patients with vestibular dysfunction complain of vertigo, difficulty balancing, and falling episodes. They may also report blurred vision on movement, such that head movements are required for balance while walking. In particular, elderly populations show a high incidence of vertigo, and may experience accidents due to the condition; thus, they should be closely monitored. Both medications and vestibular rehabilitation can improve symptoms. Although temporary symptomatic improvement may be expected in patients treated with medication, vertigo may recur and medication can delay central compensation.

In contrast, vestibular rehabilitation exploits central mechanisms of neuroplasticity (adaptation, habituation, and substitution) to increase static and dynamic postural stability and enhance visual-vestibular interactions in situations that generate conflicting sensory information [2-4]. Moreover, vestibular rehabilitation may improve static and dynamic balance and gait, thus reducing symptoms of dizziness and comorbid depression and anxiety; this ultimately increases the self-confidence and quality of life of affected patients [5]. Vestibular rehabilitation is also regarded as a safe and effective treatment modality for patients with vestibular disorders. Previous researches have shown that patients referred to vestibular rehabilitation programs at an early juncture show less disability and better gait performance after vestibular deafferentation [6-8].

However, vestibular rehabilitation may be compromised by incorrect performance of exercises, and there is a need for active effort and interest from the patient during rehabilitation; thus, efficacy varies among patients. To overcome these limitations, VR (Virtual Reality) has recently attracted attention. VR systems enable real-time simulations, have interactive functions and game-like features that promote mechanisms of neuroplasticity (adaptation, habituation, and substitution), and can improve patient motivation during vestibular rehabilitation [9]. HMD (Head Mounted displays) that provide a fully immersive experience are commonly used clinically because of their affordability and portability [10]. Also, HMDs provide high-resolution images that mimic the user’s movements, and thus give rise to an immersive experience [11].

Many previous researches have indicated that VR can improve the likelihood of successful clinical outcomes [12-15]. Additionally, VR enables visual-vestibular interactions due to abundant visual stimuli, and provides an optimized environment for rehabilitation [16]. Saprer et al. [17] reported that use of the Wii Fit Plus device during the first 2 weeks after acute vestibular neuritis was more effective in improving balance compared with placebo. However, the efficacy of VR-based vestibular rehabilitation remains controversial [18]. Clinical optimization of immersive wearable devices, such as HMDs, for dizzy patients is still required. Meldrum et al. reported that VR-based balance exercises performed during vestibular rehabilitation were not superior to conventional balance exercises during vestibular rehabilitation, but may constitute a more enjoyable modality for retraining balance after unilateral peripheral vestibular loss. VR-based vestibular rehabilitation outcomes may differ according to patient factors such as concentration levels and task accuracy. Park et al. showed that eye tracking algorithms can enhance vestibular rehabilitation using HMDs.

Here, we developed a VR-based vestibular rehabilitation program with an eye-tracking based algorithm created using the Unity3D engine. The purpose of this paper was to determine the utility of eye-tracking in vestibular rehabilitation, and to analyze the correlation between patient condition and VR-vestibular rehabilitation program.

2. Related Work

Vestibular rehabilitation training was developed in the 1940s to rehabilitate patients with brain damage and dizziness. After vestibulopathy, we observed that recovery was rapid when head movement was substantial; thus, we designed a rehabilitation method based on this principle [19-20]. In a comparison of vestibular suppressant treatment, general exercise, and vestibular rehabilitation exercise groups, the rehabilitation exercise group experienced the greatest symptom relief; in the other experimental groups, treatment was either ineffective or associated with worse outcomes [21].

The goal of vestibular rehabilitation training is to improve balance ability and vestibular function via adaptation and strengthening. The Cawthorne–Cooksey exercise is a commonly used rehabilitation exercise that is easy to implement and inexpensive. In this research, we used three rehabilitation exercises that can readily be performed in a VR environment. In addition, a vestibular adaptation exercise resulting in persistent and severe dizziness was used to habituate the patients to the sensation of dizziness; it was performed under the guidance of a specialist and patients were continuously monitored. A typical vestibular adaptation exercise is Herdman’s exercise [22] and other exercises include the ball exercise and walking in a straight line..

2.1. Cawthorne – Cooksey Training

Cawthorne – Cooksey training is shown in Fig. 1 and Fig. 2. The Cawthorne–Cooksey exercise can be performed in a lying, sitting, or standing position, and during movement; the lying position can be omitted from rehabilitation training. This research excluded difficult exercises for patients with HMD because these should be done in consultation with dizziness specialists.

Fig. 1. Cawthorne – Cooksey Training #1

a) Eye moves left and right

b) Eye moves down and up

Fig. 2. Cawthorne – Cooksey Training #2

a) Head moves down and up

b) Head moves left and right rotating head movement

2.2. Herdman training

Fig. 3 shows how to exercise Herdman training.

Fig. 3. Herdman training

a) Target should stop, eye should focus the traget when head move left, right, up, and down

b) Target should move to left, right, up and down. Eyes must focus on a target but head should be opposite direction with target.

2.3 Effect of vestibular rehabilitation training

In a research by Cohen et al. [23], individualized vestibular rehabilitation training resulted in better outcomes than general training, regardless of age, gender, and training duration. Rhee et al. [24] conducted vestibular rehabilitation training at Dankook University Hospital; patients with bilateral or central lesions showed significant improvements.

2.4 VR technology

Using VR technology, virtual environments and objects can be created and interacted with via the visual, auditory, and tactile senses [25]. This approach has fewer spatial and physical constraints than personal computer-based experimental paradigms, thus enhancing immersion [26-27]. We used the Unity physics engine to create the VR content used in this research. Unity3D can be used on multiple platforms and has a highly intuitive graphical user interface [28-29]. In this paper, we used an eye-tracking-enabled FOVE HMD which provides the frame rate was high to prevent dizziness. Table 1 shows the specification of recent HMDs.

Table 1. Specifications of recent HMDs

The user interface may not work properly when the eyeball is not tracked accurately such that, when the user wears the HMD, the eye is not measured; in this situation, the user interface is operated based on the most recently obtained left and right eye coordinates. The mean values are regarded as the “central coordinates”, according to which motion is tracked. For the HMD we tested on i7-3770K CPU with 24.0GB RAM and use NVIDIA GeForce GTX 1060 6GB. To use FOVE HMD software version of FOVE is 0.14.1 and made contents in Unity 2017.1.1.f1.

3. Implementation of vestibular rehabilitation training contents using VR

3.1 Subjects

This research was approved by our Institutional Review Board in by the ethics guidelines established by the Institutional Review Board of Soonchunhyang University Schools of Medicine. Our research included 10 healthy volunteers and 9 patients with unilateral vestibular dysfunction. The healthy volunteers recruited in this research were between 26 to 38 years of age with no reports of middle ear pathology and had never suffered from dizziness, vertigo, and imbalance. The patients in the research were recruited from the ‘vestibular clinic’ of otorhinolaryngology department of our school hospital. Patients who complained of vertigo or dizziness more than two weeks and showed unilateral vestibular dysfunction using an electro-nystagmogram were enrolled. The inclusion criteria in our research were 1) age between 20 to 70 years, 2) a clinical diagnosis of unilateral vestibular dysfunction, and 3) have complaints including imbalance, and vertigo/dizziness more than 2 weeks. Exclusion criteria included the following: 1) they were unwilling to use the VR-vestibular rehabilitation system, 2) age < 19 years, 3) any central nervous system disorder, 4) vestibular depressant medication, and 5) use of assistive devices for ambulation. All research participant was provided with an information regarding the study.

3.2 Intervention

Patients were trained on the VR-based vestibular rehabilitation system for 20–30 min while seated on a chair. All patients completed a single VR-based vestibular rehabilitation exercise. For each of the three exercises (i.e., “games”), participants started on the lowest difficulty 831 level; an adaptive algorithm automatically changed the difficulty level. If participants complained of headache or any other symptom related to the training, it was immediately stopped. The healthy subjects were tested to obtain normative game score data. They performed a single exercise, as per the patient group. For all participants, eye movement accuracy and scores were obtained. Fig. 4. shows the algorithm for proposed VR-based vestibular rehabilitation system.

Fig. 4. Flowchart of algorithm for VR-based vestibular rehabilitation system

3.3 Outcome measures

During the contents, the sight of user should follow at the movement of target via their eye movements. When the sight of user hit a target, the target changed to a different color and the user gained points; the final score was the outcome measure.

\(\text { Score }=\frac{\text { Successful hit time }}{\text { Play time }} * 100\) Weight Score = Score ∗ Speed of Object

For the eye movement we record the eye-tracking data, vector value of eye, gaze value of eye and position of eye. The data saved for 70hz and it means that record the data 70 data in 1 second and we made a graph for it. The graph’s number of x-axis and y-axis doesn’t have proper unit like mm or nm. Because we implemented the contents in Unity3D and it doesn’t have exact unit.

3.3 VR-vestibular rehabilitation system based on Eye-tracking

3.3.1 Eye movement exercise: the proposed vestibular rehabilitation content I

During the proposed vestibular rehabilitation content I, the head should be fixed and a target should be followed with the eye. The target moved leftward and rightward. The FOVE HMD can detect the eye’s original position, the eye vector, the gaze direction, and eye blinks, as well as head position and movement. A circle (“aim”) was generated to represent the movement of the eye: when the aim touched a target, the user received points and scores were weighted by level (100 points on level 1 difficulty was thus not the same as 100 points on level 3, for example). When the user moved their head in any direction, a warning message was displayed in the middle of the field of view. Fig. 5 explains the training process of proposed vestibular rehabilitation content I for Cawthorne–Cooksey training

Fig. 5. The proposed vestibular rehabilitation content I for Cawthorne–Cooksey training

3.3.1 Eye movement exercise: the proposed vestibular rehabilitation content II

The proposed vestibular rehabilitation content II also required the head to be fixed and the target to be followed with the eye; the only difference from content I was that the target moved upward and downward. Fig. 6 shows the training process of proposed vestibular rehabilitation content II for Cawthorne–Cooksey training.

Fig. 6. The proposed vestibular rehabilitation content II for Cawthorne–Cooksey training

3.3.2 Eye & Head movement exercise: the proposed vestibular rehabilitation content III

During the proposed vestibular rehabilitation content III, targets (“meteors”) moved in random directions, and at random distances from the user. The user was required to follow the targets with their eyes and thus fire “bullets” at them, similar to a “first-person shooter” game. Fig. 7 shows the proposed vestibular rehabilitation content III for Herdman training with first-person shooter type.

Fig. 7. The proposed vestibular rehabilitation content III for Herdman training

3.4 Statistical analysis

Differences in scores (minimum, maximum, and average) between the healthy control and patient groups were analyzed for the three exercises, and scatter plots were generated to graph these differences. We recorded eye movements at 70 frames per second and omitted blink data. A p-value < 0.05 was considered significant. The analyses were performed using SPSS for Windows software ([1.0.0.1298]; SPSS Inc., Chicago, IL, USA).

4. Experimental Classification Results and Analysis

The usability of the VR-based vestibular rehabilitation system was evaluated in 9 patients with unilateral vestibular dysfunction and 10 healthy subjects in a single testing session. All participants that satisfied the inclusion criteria received instructions in the VR-based vestibular rehabilitation exercises during a 20-min session and then completed a full 20–30 min training session. Table 2 shows the characteristics of the healthy controls and patients; the average age was 30.7±3.8 and 52.0±15.3, respectively. The gender ratios were 5:5 in the healthy control group and 4:5 in the patient group. Three patients had a diagnosis of benign paroxysmal positional vertigo, two had vascular necrosis, one had vascular necrosis, one had Meniere’s disease, and two exhibited vestibular dysfunction. For healthy group the vestibular are not affected by age factor and both healthy control group and patient group are for comparison with score and accuracy during the test.

Table 2. Characteristics of participants

Table 3 and Fig. 8(A) showed the average of game content I score in the healthy and patient groups. Depending on the difficulty of game content I, the healthy group showed 31.58±0.40 in the level l, 29.24±0.84 in the level 2 and 28.66±0.60 in the level 3, and the patient group showed 21.85±2.66 in the level 1, 18.05±1.95 in the level 2, 16.29±2.37 in the level 3, respectively. The healthy group scored almost 10 points higher than the patient group depending on the different level and these differences in content I were statistically significant (p=0.001, p=0.000, and p=0.000). Additionally, Table 3 and Fig. 8(B) showed the eye-movement accuracy in the healthy and patient groups. Depending on the difficulty of game content I the healthy group showed 94.32±1.27 in the level l, 88.27±2.36 in the level 2 835 and 90.06±1.90 in the level 3, and the patient group showed 65.99±8.04 in the level 1, 54.32±6.04 in the level 2, 50.20±7.39 in the level 3, respectively. The eye-movement accuracy of healthy group in content I showed higher than the patient group depending on the different level and these differences were statistically significant (p=0.002, p=0.000, and p=0.000).

Table 3. Average of game score and eye-movement accuracy in the proposed vestibular rehabilitation content I

Fig. 8. Comparison of game score and eye-movement accuracy in the proposed vestibular rehabilitation content I

(a) game score (b) eye-movement accuracy

Table 4 and Fig. 9(A) showed the average of game content II score in the healthy and patient groups. Depending on the difficulty of game content II the healthy group showed 30.83±0.45 in the level l, 30.33±0.45 in the level 2 and 29.58±0.82 in the level 3, and the patient group showed 23.70±1.25 in the level 1, 23.05±1.12 in the level 2, 21.39±0.50 in the level 3, respectively. The healthy group scored about 8 points higher than the patient group depending on the different level and these differences in content II were statistically significant (p=0.000, p=0.000, and p=0.000). Additionally, Table 4 and Fig. 9(B) showed the eye-movement accuracy in the healthy and patient groups. Depending on the difficulty of game content I the healthy group showed 92.77±1.47 in the level l, 91.77±1.49 in the level 2 and 92.07±1.59 in the level 3, and the patient group showed 72.27±3.75 in the level 1, 69.31±3.63 in the level 2, 67.23±1.98 in the level 3, respectively. The eye-movement accuracy of healthy group in content II showed higher than the patient group depending on the different level and these differences were statistically significant (p=0.002, p=0.000, and p=0.000).

Table 4. Average of game score and eye-movement accuracy in the proposed vestibular rehabilitation content II

Fig. 9. Comparison of the proposed vestibular rehabilitation content II in healthy and patient groups

Table 5 and Fig. 10 showed the average of game content III score in the healthy and patient groups. The healthy group showed 97.50±1.18, and the patient group showed 79.11±7.30 respectively. The healthy group scored 18 points higher than the patient group these differences were statistically significant (p=0.018).

Table 5. Average of game score for the proposed vestibular rehabilitation content III

Fig. 10. Comparison of the proposed vestibular rehabilitation content III in healthy and patient groups

Fig. 11 showed eye-movement in the healthy and patient group. The eye-movement in healthy group was more concentrated than the patient group. The results implied that eye-movement in vestibular rehabilitation could be used as an indicator depending on the patient’s condition or disease.

Fig. 11. Scatergram about eye-movement during the proposed vestibular rehabilitation content I and II

5. Conclusion

In this paper, VR-based vestibular rehabilitation exercises created using the Unity3D engine were tested with healthy controls and patients. A limitation of such rehabilitation training is that stimuli alerting the user to the need for concentration cannot be provided. Moreover, space and financial restrictions may limit applicability, and falls and collision can occur during training. However, VR-based rehabilitation training provides an interesting and immersive environment, which may improve patient interest, and remains much safer than conventional rehabilitation training. The use of an eye tracking-enabled HMD can increase data accuracy; HMDs without eye tracking capabilities can only detect head direction and degree of tilt, 838 whereas newer models can track progress in real-time and provide feedback to the patient. As VR-based rehabilitation training becomes more realistic, we expect to further improve the rehabilitation exercises developed in this research. Furthermore, we plan to get more data from more people to make specific relation between our exercises and patient’s condition of their symptom. In that case the data can have high quality such as accuracy and high reliability, that would be the first step of home training rehabilitation which prove that value.