Virtual reality (VR) mental health therapy has been studied extensively, and some of the resulting interventions are already used in clinical practice. However, high-quality VR systems are still relatively expensive. There are some indications that a cheaper, low-quality VR system produces di erent e ects in its users, though studies remain scarce. We theorize that the e ectiveness of a VR intervention is determined partly by how well it directs attention towards or away from the users real body. We therefore propose a study comparing how image-quality and biofeedback during a breathing task a ect anxiety levels. We hypothesize that high image quality distracts from the task, making the intervention less e ective. Conversely, placing biofeedback in the virtual environment could direct attention back to the users body, increasing effectiveness. Initial results suggest individual di erences in how attention is managed. A full-scale study could inform a more structured, attentionbased design approach towards creating VR interventions.
1.1
Over the last couple of decades, entertainment based health interventions have
been studied extensively and some are already successfully used in clinical
practice [
Immersive virtual reality, where the user is viewing the virtual world through
a head mounted display, can induce a strong sense of presence - the feeling of
actually entering the computer generated world [
VR can be used as a diagnostic tool, as treatment and as resilience training
[
However, although the cost associated with VR has fallen dramatically in the
past couple of decades, even commercially available immersive VR systems are
still relatively expensive, making them inaccessible for everyday use [
The accessibility of a VR system is mostly connected to its price. But before
the use of cheaper systems is encouraged, we need to know whether these
systems di er in their performance from their more expensive counterparts, and in
what way. Research on the di erence between higher and lower quality VR Head
Mounted Displays (HMD's) is unfortunately scarce. Ho man et al. [
As VR technology is advancing rapidly, recent studies make use of cheaper
VR systems more often than before, often showing that there is big potential
in developing smart-phone based VR health applications [
The direction of attention is one mechanism that seems to strongly in uence the
e ectiveness of the VR intervention. We theorize that di erent types of virtual
reality interventions have di erent goals as to the direction of the attention of
the user. For instance, in pain relief the goal is to distract the user from their own
body and the pain it is experiencing [
There are multiple ways of reducing feelings of anxiety and stress, one of
which is through deep, diaphragmatic, slow breathing techniques [
The creation of an interesting and engaging VR experience that is
nonetheless not too distracting is a problem that has been approached several times
through the addition of biofeedback [
Redirecting the users attention away from the real body and unto a complex, high-quality virtual environment might be useful in pain relief, but is likely not ideal for acquiring deep breathing techniques and reducing anxiety. In these cases, a low-quality VR environment might su ce or even work better. However, adding biofeedback visualizations to the virtual environment might bring the attention back to the body, regardless of how engaging the virtual world itself is. We therefore propose to study both the image-quality and the presence of biofeedback in the context of a virtual breathing exercise to reduce anxiety.
We hypothesize that a higher quality VR environment with biofeedback visualizations will results in the highest reduction of anxiety levels, followed by a lower quality VR with biofeedback. We expect the high-quality VR experience to draw more of the users attention, which is then directed towards their breathing via the biofeedback. We hypothesize that both non-biofeedback conditions will perform poorly, with the higher quality VR being the worst. We expect that the high quality virtual environment would distract the user more, making them less focused on their body and breathing and thus reducing the e ects of the intervention.
By studying the mechanisms behind the di erent e ects of image-quality and the presence of biofeedback, we hope to gain more knowledge on how to design future VR health care applications and develop the basis of design practices that can make them more accessible.
12 participants (5 female, 7 male) were recruited on the campus of Tilburg University. For this initial exploratory pilot, participants were not selected speci cally for anxiety disorders.
There are four conditions: high quality VR with biofeedback, high quality VR without biofeedback, low quality VR with biofeedback and low quality VR without biofeedback. Each participant is assigned to two conditions; either high quality VR with and without biofeedback or low quality VR with and without biofeedback, making this study a mixed 2x2 design with the image-quality as the between subjects variable and the presence of biofeedback as the within-subject variable. The order of the conditions is randomized and counterbalanced over the participants. 2.1
Anxiety Questionnaires Level of anxiousness is measured with two di erent
kinds of questionnaires. At the beginning of the experiment the participant are
asked to ll in the Trait Anxiety Inventory, with items such as I am content; I
am a steady person. [
State anxiety is measured with the use of Subjective Units of Discomfort
Scale, or SUDS [
Anxiety Induction In between the two conditions we lower the induced
relaxation to avoid carry-over from the rst condition into the second. To achieve
this we use the Velten Mood Induction Procedure - the participants read and
try to imagine a number of negatively loaded sentences for one minute [
Our virtual environment is created with Unity 2017.2.0f3. and looks like a calm forest. This forest was made with the 'FantasyEnvironments' asset in Unity. The participant sits in the virtual forest at one predetermined spot. A total of 13 randomly placed lights are visible.
In the conditions with biofeedback, the intensity of these lights re ects the participants diaphragm expansion through a connection between the Arduino and Unity, using the ARDUnity 1.0.8 asset. Breathing in correctly brightens the lights, and breathing out dims them. In the conditions without biofeedback the lights are o .
The head mounted display used in this experiment is the HTC Vive (2160x1200 pixels max, 90Hz, 110-degree eld of view). We created a high and a low quality image within this same HDM by adjusting quality aspects within Unity, speci cally the texture quality, shadow resolution, and density of details in the environment (see gure 1 and 2). 2.3
Informed consent is obtained from all participants. Participants then receive a short instruction on deep, diaphragmatic breathing techniques and practice these shortly. Then the participant lls in the Trait Anxiety Inventory, puts on the stretch-sensor band and read the Velten Mood Induction sentences. The participant then rates the SUDS, the VR HMD is placed on the head of the participant and the rst condition starts. Near the end of virtual experience the participant is asked to rate the SUDS again, followed by the removal of the HMD. The participant now again rates the SUDS and take a short break. Then, the same procedure is followed for condition two. After both the conditions the participant is debriefed.
During this exploratory pilot study, several trends were observed. Most importantly, there seems to be a wide range in individual responses to attentionmanagement within virtual reality. While some participants indicated that they found the biofeedback lights calming and fun, others found them to be distracting. In general, if a participant did not achieve the correct deep breathing patterns, they tended to describe the lights as random, in one case even as annoyingly so. There were also some reports of novelty bias, were participants who hadn't experienced VR before reported feeling excited or nervous for that reason. This could potentially cloud the measurements.
SUDS scores tended to decrease during all virtual reality experiences, although this e ect was not signi cant. In the non-biofeedback conditions, the SUDS score was slightly higher during the high-quality condition compared to lower quality, but this di erence was also not signi cant. For these measurements, a full-scale study will hopefully bring more clarity.
Strong deep breathing patterns were very clearly observed in the diaphragm expansion data from several participants. In gure 3 you see an example of a participant who maintained deep breathing throughout the experiment, whereas 4 shows the patterns of a participant who self-reported that they didn't manage to achieve deep breathing. There were no strong trends observed in the deep breathing data related to the di erent conditions. There seem to be quite strong individual di erences, so a large full-scale study might identify potential patterns more e ectively. The use of relatively new technologies and media in cognitive science comes with several challenges pertaining to experimental design. We will recount a few of our considerations here.
Firstly, the use of questionnaires could be detrimental to a users sense of presence in the virtual environment. It becomes more di cult to switch between virtual and non-virtual tasks, and could have unforeseen e ects. Tools like the SUDS scale allow for a shorter assessment and verbal answer, allowing the user to maintain their sense of presence in the virtual world.
Secondly, the biofeedback used (breathing) can be measured in a variety of ways. One especially non-invasive option we explored was a temperature sensor under the nose - warm air signi es breathing out, cold air breathing in. We found however that the sensor did not responds quickly enough to the temperature changes. Another option would be to work with a microphone. This has the added advantage that it would be relatively easy to use with a smart-phone HMD, as the microphone is already a part of the device. However, this does not allow you to distinguish between deep diaphragmatic breathing and shallow chest breathing. The stretch sensor band is slightly constricting around the users abdomen, but does allow us to use biofeedback only when the user is breathing deeply. We constructed the stretch sensor ourselves using conductive thread, yarn and elastic band. This, together with the use of the Arduino micro-controller, results in a relatively a ordable and accessible, yet reliable deep breathing sensor. 3.3
Should the full-scale version of this study show that not only high but also low quality VR indeed reduces anxiety signi cantly, it would be interesting to continue testing with actual smart-phones, preferably with low-priced accessible HMD's such as the Google Cardboard. Future research could additionally examine the in uences of biofeedback lag, a smaller eld-of-view, and alternative biofeedback sensors on the reduction of anxiety.
If our study shows that the way in which attention is redirected within the virtual world has an e ect on anxiety reduction, the next logical step would be to repeat the study with an intervention that has a di erent goal in mind, such as directing attention away from the body. For example, one would hypothesize that for pain relief environments, a higher quality VR experience would work best, while adding biofeedback would negate the e ects. In other words, the exact opposite result from the current study.
Another interesting avenue is the way biofeedback is placed within the virtual environment. Instead of the biofeedback visualization a ecting the external virtual environment, it would be interesting to look at the e ect of placing the biofeedback on the users virtual body. This could direct the users attention to their body even stronger. Alternatively, the biofeedback could be perceptualized in a way other than visually, for instance via audio.
If we elucidate the mechanisms behind what makes virtual reality health care intervention work in di erent scenarios, it could give us more information on which interventions can easily be made accessible and in which way. For instance, breathing exercises might be suitable to use with low-quality devices, while pain relief experiences might not. Giving the patient the possibility to experience the treatment wherever and whenever they want enormously increases their freedom and autonomy.
The development of the technology itself also means that these outcomes will continue to shift. The image-quality and processing capabilities of smart-phones is increasing every year. Rather than wait until smart-phone based HMD's have the same speci cations as the high quality HMD's of today, we believe well researched, accessible VR based health applications can already be created in the coming years.