Exploring Biofeedback with a Tangible Interface Designed for Relaxation

Morgane Hamon, Rémy Ramadour and Jérémy Frey

Ullo, 40 rue Chef de Baie, La Rochelle, France

Keywords:

Biofeedback, Tangible Interface, Relaxation, Ambient Display, Interoception.

Abstract:

Anxiety is a common health issue that can occur throughout one’s existence. In this pilot study we explore an

alternative technique to regulate it: biofeedback. The long-term objective is to offer an ecological device that

could help people cope with anxiety, by exposing their inner state in a comprehensive manner. We propose

a ﬁrst iteration of this device, “Inner Flower”, that uses heart rate to adapt a breathing guide to the user; and

we investigate its efﬁciency and usability. Traditionally, such device requires user’s full attention. We propose

an ambient modality during which the device operates in the peripheral vision. Beside comparing “Ambient”

and “Focus” conditions, we also compare the biofeedback with a sham feedback (ﬁxed breathing guide). We

found that the Focus group demonstrated higher relaxation and performance on a cognitive task (N-back).

However, there was no noticeable effect of the Ambient feedback, and the biofeedback condition did not yield

any signiﬁcant difference when compared to the sham feedback. These results, while promising, highlight

the pitfalls of any research related to biofeedback, where it is difﬁcult to fully comprehend the underlying

mechanisms of such technique.

1 INTRODUCTION

At times, to endure fear might be beneﬁcial, when

one is facing dangerous situations or unknown events.

“Fight or ﬂight” responses were proven important for

survival. However if stress becomes chronic and is

not treated, it can be a factor of sleep disorders or car-

diovascular disease (Kivimäki et al., 2002). It can also

lead to a pathological state of anxiety when the anti-

cipation of stressing stimuli is sufﬁcient to trigger the

same symptoms as with the actual appearance of sti-

muli. Finding effective and lasting solutions to reduce

stress is necessary to alleviate this public health pro-

blem, which impedes the lives of many. Treatments

exist against stress and anxiety, but they might re-

quire a strong and timely involvement (i.e. therapies)

or provoke side effects (i.e. drugs). Studying anx-

iety and offering alternative solutions to drugs (sport,

yoga, mindfulness) is a growing body of research, and

nowadays tools exist to let people autonomously re-

ﬂect on their states and better act upon themselves.

1.1 Biofeedback

Biofeedback is a method that enables users to learn to

control autonomous bodily processes. Biofeedback is

part of a larger notion known as "interoception" which

Figure 1: In this example heart-rate is measured through

a smartwatch; tangible interfaces serve both as breathing

guides for the user and as biofeedback devices (image



Inria, photograph C. Morel).

is deﬁned by (Farb et al., 2015) as "the sense of sig-

nals originating within the body" or "the process of

receiving, accessing and appraising internal bodily

signals". Biofeedback relies on physiological measu-

rements and is getting more and more popular thanks

to the increasing availability of non-invasive sensors

(Figure 1). Most of the time, signals originate from

respiration, electrodermal activity (EDA) or heart rate

Hamon, M., Ramadour, R. and Frey, J.

Exploring Biofeedback with a Tangible Interface Designed for Relaxation.

DOI: 10.5220/0006961200540063

In Proceedings of the 5th International Conference on Physiological Computing Systems (PhyCS 2018), pages 54-63

ISBN: 978-989-758-329-2

(HR) (McKee, 2008).

While these processes are mostly involuntary and

with weak level of consciousness, through physiolo-

gical sensors it becomes possible to visualize them in

real-time. Users are then able to view effects of a par-

ticular behavior on their physiology . For example if

a person wearing a HR sensor begins to run, HR is

going to increase suddenly. Connected to a biofeed-

back display the increase of HR could be symbolized

by a color change, a blinking light, or simply seen as

a graph. This information can then be used to help

users regulate their state, for example stay within a

speciﬁc HR range in to prevent injuries or maximize

the outcomes of the training. Biofeedback applicati-

ons are diverse, from rehabilitation to stress manage-

ment (McKee, 2008).

It is worth noting that despite the fact that bio-

feedback has been investigated and applied for deca-

des, its effectiveness has not been systematically de-

monstrated. Notably, even though a review such as

(Yucha and Montgomery, 2008) presents many stu-

dies across a variety of medical applications, there

are hardly any comparisons between biofeedback and

sham feedback, which is the only way to form a pro-

per control group and pin-point the real efﬁcacy of the

technique.

Still, as a drug-free and a non-invasive approach,

potential risks and side effects of Biofeedback are

small. With a medical opinion it could be a compro-

mise for people who can not or do not want to take

drugs (e.g pregnant women).

1.2 Heart Rate Variability: A Marker

of Anxiety

One of the most studied case of biofeedback for stress

management is the cardiac activity. Heart Rate is un-

der control of the autonomous nervous system (ANS)

and is not constant at rest. Heart Rate Variability

(HRV) – a marker of the evolution of HR over time

– is a representative index of ANS activity. Among

the variety of factors that could inﬂuence HR, a low

HRV has been shown to be correlated with impaired

parasympathic activity, higher anxiety, and a variety

of disorders (Vaschillo et al., 2006). (Tan et al., 2011)

showed that veterans with Post Traumatic Stress Dis-

order (PTSD) had a HRV signiﬁcantly lower than sub-

jects without PTSD. The HRV was compared between

veterans with PTSD who received HRV biofeedback

in addition to their regular treatment and veterans wit-

hout the additional biofeedback treatment. The re-

sults indicated that the group with HRV biofeedback

had signiﬁcantly increased their HRV while reducing

symptoms of PTSD compared to the other group. Fin-

ding ways to assist people to learn to increase their

own HRV and improve their health is one of the mo-

tivations of our study.

1.3 Increasing Heart Rate Variablity

Deep breathing is a well documented manner to incre-

ase the HRV. “Cardiac Coherence” is a notion accor-

ding to which respiration and cardiovascular functi-

ons are synchronized. That means the HR increases

during the inhalation and decreases during exhalation,

and so periodically (McCraty et al., 2009). Authors

proposed that 6 breaths per minute is the respiratory

frequency that allows one to reach cardiac coherence

and the highest amplitude in heart rate oscillations

(hence the highest HRV) (DeBoer et al., 1987). Car-

diac Coherence is then a priori obtained when brea-

thing at a 0.1Hz frequency which corresponds to six

10-second breathes per minute.

Several studies have investigated a static breathing

guidance at 6 breaths per minute to reduce stress.

For example in (Yu et al., 2015), where authors sho-

wed that HRV was higher after the breathing exer-

cise, while subjective impressions – as measured by

the State Trait Anxiety Index questionnaire, or STAI

(Spielberger et al., 1970) – were not different. In

(Dijk and Weffers, 2011) authors designed an im-

mersive system which is composed of a blanket con-

taining small vibrating motors (haptic stimulus) and

headphones (audio stimulus). Haptic and audio sti-

muli are synchronized and generated at such a fre-

quency that the user can follow as a breathing gui-

dance. One of the studied frequency modality was

6 breaths/min. Some participants reported the fre-

quency as too fast or too slow. The Authors highlig-

hted that a poorly adapted frequency can potentially

create hypo/hyperventilation.

Even if a static guidance might, on average be op-

timal for the population, it is not the best way to max-

imize HRV for each individual. As a matter of fact,

(Vaschillo et al., 2006) found that resonant frequency

(equivalent to the cardiac coherence) differs accor-

ding to each person. By exposing a breathing gui-

dance from 4.5 to 6.5 breaths per minute to their par-

ticipants, they managed to deﬁne the best frequency

for each subject.

As such, we opted for an adapted breathing gui-

dance to create our biofeedback; in our current study

we compare it with a ﬁxed breathing feedback (i.e.

“Dynamic” vs “Static” feedback) in order to have a

better grasp on the effect of an adapted biofeedback.

Exploring Biofeedback with a Tangible Interface Designed for Relaxation

1.4 Shaping the Best Biofeedback

(Muench, 2008) proposed a certain type of biofeed-

back to achieve higher HRV and reach a resonant fre-

quency. In this study, a graph based on HR ﬂuctua-

tions was created, having users inhaling until the HR

reaches a “peak” and exhaling until it reaches a “pit”.

Because of the improved HRV as compared to bre-

athing exercises agnostic to HR, we also designed a

biofeedback device that would use HR measurements

to propose an adapted breathing guide.

However, the feedback modality used to convey

the information back to the user is important, and

graphs are not the only way to present a breathing

guidance. Actually, such kind of feedback might even

impede acceptability because it could appear as being

too judgmental due to its close relationship to a me-

tric. In order to craft a more “organic” and yet infor-

mative biofeedback, we decided instead to rely on a

physical object to present the feedback.

Indeed, thanks to recent advances in human-

computer interaction, is is now possible to directly

integrate digital information within users’ surroun-

dings, for example through the use of tangible inter-

faces. These interfaces have been proven effective to

help people learn about bodily processes – see e.g.

(Fleck et al., 2018). Through tangible interfaces, bi-

ofeedback can then be more easily integrated in the

natural settings of users, and become part of a speci-

ﬁc scenario.

1.5 Ambient Feedback

Ambient computing contrasts with disrupting notiﬁ-

cations. In this context “ambient” refers to informa-

tion that is being presented in the peripheral attention

of users (MacLean, 2009). Ambient devices do not

mobilize attention. They require minimal efforts from

the user and yet they provide informative feedback;

rather than “pushing” a notiﬁcation it is up to users to

“pull” information when they require it.

Along those lines, (Moraveji et al., 2011) high-

lights two ways to train respiration: consciously thin-

king about it or learning to follow an external stimulus

such as a pacing light or an auditory guide. While the

former method implies focus of attention, the latter, if

proven effective, could alleviate the required amount

of cognitive resources (Hazlewood et al., 2011).

In (Moraveji et al., 2011), authors investigated

whether a breathing guidance at the periphery of the

screen during work had a inﬂuence on breath rate. As

this type of guide does not require full attention they

called it Peripheral Paced Respiration (PPR). This

ambient guide can be very useful by allowing the user

to fully commit to another task. The guide rate setting

was 20% below the user baseline. They showed signi-

ﬁcant difference on breath rate depending on the acti-

vation of the PPR. The same thinking led (Azevedo

et al., 2017) to develop a wearable device that deli-

vers tiny vibrations on the wrist with a frequency 20%

slower than the participant resting HR. Users didn’t

know the function of the device and were preparing an

oral presentation (to induce stress). Results showed

that the control group had signiﬁcantly higher anxiety

according to questionnaires (STAI) and physiological

data (EDA).

(Schein et al., 2001) investigated in a longitudinal

study if a device called BIM (for Breathe with Inte-

ractive Music) had a positive inﬂuence on Blood Pres-

sure (BP). The BIM device creates a musical pattern

which is related to the user breathing rate. A 10 minu-

tes long quiet synthesized music recording was used

as an active control. Results showed BP reduction

was greater in the experimental group compared to

the control, and seem to have a long-term effect (sig-

niﬁcantly different 6 months after).

Based on these various ﬁndings, we decided to in-

vestigate not only a “Focus” but an “Ambient” utiliza-

tion of a tangible biofeedback device as well. In order

to better frame our experimental design, we turned

to previous work from psychology and physiological

computing that aimed at investigating various dimen-

sions of stress.

1.6 Evaluating and Inducing Stress

Inducing stress to evaluate short-term effects of bio-

feedback devices is common in the literature. To arti-

ﬁcially put participants in a more stressed state incre-

ases the range of measurements and help to reduce va-

riability between subjects. Different type of stress can

be induced, physical (e.g. extreme heat or cold), psy-

chological (e.g. increase in mental workload) or psy-

chosocial (e.g. public speaking) (Mühl et al., 2014).

Psychosocial stress can be induced by faking an in-

terview – Trier Social Stress Test (Kirschbaum et al.,

1993) – or simply asking participants to prepare a

speech, as in (Azevedo et al., 2017).

In (Yu et al., 2015) authors induced psychologi-

cal stress with a mathematical task for a ten-minutes

period. Several manners exists to check if the stress

inducing task had an effect. First, it can be detected by

recording physiological data, for example with HRV

as a marker of anxiety. EDA is also a common re-

corded measure – e.g. (Roseway et al., 2015) – to

detect arousal. Because physiological signals might

have poor speciﬁcity, those indicators do not repre-

sent how participants are feeling, what is their state

PhyCS 2018 - 5th International Conference on Physiological Computing Systems

of mind. Questionnaires can compensate for this si-

tuation. To measure anxiety the most frequently used

questionnaire is the STAI as in (Yu et al., 2015), (Aze-

vedo et al., 2017), or (Mühl et al., 2014). In the latter

work authors compared the robustness of various phy-

siological signals to detect stress. They crossed psy-

chosocial stress (TSS) with psychological stress, by

manipulating the amount of cognitive workload en-

dured by participants. To do so, they employed the

N-back task, which leverage on memory load (Owen

et al., 2005). We chose to use the N-back task as well

to induce stress in our study since it is effective in

doing so and since we can easily compute a perfor-

mance metric to sense how participants might be af-

fected by the exposition to a tangible biofeedback.

1.7 Objectives

In the present study, our ﬁrst objective is to investigate

if an ambient modality could have positive effects on

stress. To explore this issue we compared two moda-

lities: Ambient vs Focus. “Ambient” stands for the

presence of the device during the whole experiment

(30min) in the peripheral visual ﬁeld of the partici-

pant. “Focus” refers to a short (6 min) utilization in

the middle of the experiment requiring full-attention.

Our second objective is to check the relevance

of breathing guidance based on HR . Indeed, as we

propose a new device it is important to consider that

any observable effect might be due to a “novelty ef-

fect”. Moreover, we wanted to assess the usefulness

of actually measuring physiological activity. Hence

we compare a true biofeedback (“Dynamic” condi-

tion, where the breathing guide is adapted using HR)

with a sham biofeedback, or “pseudofeedback” (“Sta-

tic” condition, where the breathing guide is set to a

ﬁxed rate). As explained by (Health et al., 1982), a

“pseudofeedback is deﬁned as a non contingent stimu-

lus presented in exactly the same manner as the true

biofeedback and with the intention of having subjects

believe that it is true biofeedback”.

2 STUDY

This study possesses a 2 (Attention: Ambient vs

Focus) × 2 (Biofeedback: Dynamic vs Static)

between-subject experimental plan. As a result

we split our participants into four distinct groups:

Ambient-Dynamic, Ambient-Static, Focus-Dynamic

and Focus-Static.

Because in the Focus group the device is used only

mid-experiment, the ﬁrst part of the experiment in Fo-

cus serves as a control group (no device) for the inves-

tigation of attention.

Our hypotheses are:

H1. Exposition to an Ambient feedback (H1a)

or to a Focus feedback (H1b) reduces psychological

stress.

H2. An adapted breathing guidance with a bio-

feedback device (Dynamic) will reduce stress as com-

pared to a pseudofeeback (Static).

H3. The usability resulting from the use of a

biofeedback device is improved as compared to a

pseudofeedback (Focus-Dynamic vs Focus-Static).

2.1 Inner Flower

Figure 2: Left: Examples of prototypes toward an tangible

biofeedback. Right: Design of the Inner Flower used in the

study.

In order to test our hypothesis, we iterated over

various form factors to craft a tangible biofeedback

that could both act as an ambient feedback and as a

breathing guide. Among the biofeedback modalities

that are traditionally employed – visual, audio or hap-

tics – (Frey et al., 2018) suggests that there is little

difference regarding the effectiveness of conveying a

breathing feedback. We chose to rely on visuals with

color LEDs since they offered many degrees of free-

dom to work with (e.g. location of the lights, inten-

sity, color, speed of the animation) and since it was

easy to implement.

We employed laser cutting, Arduino-based micro-

contollers (Arduino mini), and Adafruit

Neopixel

LEDs to prototype several artifacts (Figure 2). It

quickly became apparent that frosted plastic was the

best material to work with in order to improve the

LEDs’ diffusion, smooth the light and prevent an im-

pression of a pixelated display, which would be too

much reminiscence of a desktop display. Regarding

https://www.adafruit.com/

Exploring Biofeedback with a Tangible Interface Designed for Relaxation

the actual appearance of the object, after having ex-

plored more abstract shapes, we decided to create a

somewhat stereotypical ﬂower in order to convey a

sense of well-being. Additionally, as one might want

to smell the scent of a ﬂower, it is in a way a valid

proxy for breathing.

Then we had to design the feedback that would

guide the breathing. While it was straightforward to

pick movements as a guidance, several adjustments

were required in order to smooth the animation of the

LEDs that were disposed on the “petals” of the ﬂower.

In the ﬁnal version (Figure 2, right), the user would

breathe in when the lights from of the Inner Flower

move outwards and breathe out when the lights move

inward. In the Static condition this animation is set

to 6 breaths/min. In the Dynamic condition, the bre-

athing rate (BR) is coupled with the heart rate of the

participant. We start our investigation with a simple

coupling: BR = HR/∆. The divider ∆ has a default

value of 15. It can be set to adapt the breathing gui-

dance to each user during a calibration phase. During

the Dynamic condition, the breathing guidance varies

in real-time according to the HR of the user so as to

reach cardiac coherence. For example with a HR at 90

beats per minute and a divider set to 12, the resulting

breathing guide pulses 7.5 times per minute.

There is no direct representation of the HR (e.g.

no light blinking at the pace of the heart rate) so as

not to overwhelm users with information not related

to current usage scenario.

2.2 Measures

During the study we were interested in collecting two

types of data: the (psychological) stress level of par-

ticipants, and a usability index related to the tangible

biofeedback. Stress was assessed along three dimen-

sions: physiological activity (HR), behavioral measu-

res (performance in a N-back task) and questionnai-

res (STAI). Usability was assessed through a questi-

onnaire (USE) administrated at the end of the experi-

ment for those groups that explicitly used the device

(i.e. Focus groups).

2.2.1 Heart Rate

Each participant was equipped with a smartwatch me-

asuring heart rate (“Link” from Mio

), placed on their

non-dominant hand. Mio smartwatchs employ pho-

toplethysmography (PPG) to compute heart-rate, a

technique which basically detects variations in skin’s

color to assess heartbeats. This solution was prefer-

red over electrocardiography (ECG) to improve com-

https://www.mioglobal.com/

fort and acceptance of the system, which is meant to

be used in ecological settings, outside the laboratory.

Even though PPG is less robust than ECG, being that

the participants are steady and seated, i.e. without the

risk of creating motion artifacts throughout this study,

PPG is a sufﬁciently good sensor. During pilot stu-

dies, we validated that the instantaneous HR measu-

red by this particular smartwatch was accurate enough

to detect changes in HRV associated with deep relax-

ation.

Data was collected over Bluetooth, processed in

real-time in the Dynamic condition and stored for

further analysis. From HR measurements we focu-

sed our investigation on one index of HRV: RMSSD

– root mean sum of the squared differences. RMSSD

takes as input the inter-beat interval (the inverse of

the instantaneous HR measured by the smartwatch).

It is one of the best indicator of cognitive workload

(Mehler et al., 2011), a speciﬁc type of psychological

stress induced during the experiment.

2.2.2 N-back

Figure 3: 2-back task. Participants have to click left when

the displayed letter is the same as two steps before.

The N-back task served both to induce psycholo-

gical (cognitive) stress, and to evaluate the cognitive

load of participants. The latter is revealed by calcu-

lating participants’ performances. During this task,

as in (Mühl et al., 2014), each letter appears on the

screen for 0.5 second every 2 seconds. Participants

have to determine whether the current letter is the

same as the one they saw N steps before (left click

on a mouse) or not (right click). We employed a 2-

back task (Figure 3), which showed to induce a high

workload level (Mühl et al., 2014). Typically, when

workload increases the performance during the task

decreases.

An immediate feedback is provided to inform par-

ticipants whether their answers are correct or not.

Two N-back tasks were presented to each participant,

denoted as “N-back1” and “N-back2”. Each task is

PhyCS 2018 - 5th International Conference on Physiological Computing Systems

comprised out of three sequences of one minute (30

letters). The ﬁrst task (N-back1) includes an additi-

onal sequence at the beginning that acts as a training

session. Over each N-back task we measured partici-

pants’ performance (percentage of correct answers).

2.2.3 STAI

The State Trait Anxiety Index (STAI) enables the me-

asurement of self-reported anxiety levels (Spielberger

et al., 1970). The STAI Y-A version of the question-

naire measures anxiety as a state (about one’s current

endeavor). The classical version is composed of 20

items (e.g. “I am tense”; “I feel content”) that parti-

cipants rate on a 4-point Likert scale. A higher score

reﬂects a higher anxiety state. Because we adminis-

trated multiple times the STAI during the experiment,

we favored a six-item short-form (Marteau and Bek-

ker, 1992). It was faster to ﬁll by participants and

still produced similar scores to those obtained with

the full-form.

2.2.4 Usability

In order to assess the usability of the device, we adap-

ted the USE questionnaire (Lund, 2001). We remo-

ved the “Usefulness” subscale because it was not rele-

vant to our study – participants did not use the device

long enough. Questions were translated to French.

The resulting questionnaire comprises three subsca-

les: “Ease of use”, “Ease of learning” and “Satis-

faction”. Each is composed of three items (e.g. “I

easily remember how to use it”, “It is simple to use”).

Participants had to state their agreement to each sen-

tence on a 4-point Likert scale (“Not at all” ... “Very

much”). Since such usability questionnaire is poorly

suited to assess an ambient device, with which users

merely interact directly and/or consciously, the USE

was only administrated to participants of the Focus

groups.

2.3 Participants

A total of 36 volunteers (18 females) took part in the

study. Overall the mean age was 23.80 years old (SD

= 4.82). The demographics per group is depicted in

Figure 4.

2.4 Procedure

The timeline of the experiment is presented Figure 5.

The experiment took place in a quiet room, deprived

of distractions, and we detail it step by step as follows.

Figure 4: Demographics of our groups.

Figure 5: Timeline of the experiment.

Upon entering the room and sitting at a table, par-

ticipants signed a consent form. Afterwards we pro-

ceeded to explain brieﬂy how the experiment would

take place and equipped participants with the smart-

watch. The Inner Flower was then switched on and its

core functionalities were explained to participants. In

the Focus-Dynamic group, a calibration phase occur-

red in order to determine a pace that would suit users

(i.e. obtaining a breathing guide that would not appear

too fast nor too slow). After the calibration, the device

was switched off. On the other hand, the Focus-Static

group had no calibration, hence the device was sim-

ply switched off. Lastly in the Ambient groups the

device was left active on the side of the table, in the

peripheral vision of participants.

Afterwards, to induce psychological stress and

control for the effect of the N-back task participants

fulﬁlled a ﬁrst STAI (STAI-1), performed a ﬁrst N-

back task (N-back1) and ﬁnally fulﬁlled a second

STAI (STAI-2). At this point of the experiment, by

comparing Ambient groups (active device on the side)

with Focus groups (devices turned off) we are able to

investigate the effect of an ambient breathing guide.

Exploring Biofeedback with a Tangible Interface Designed for Relaxation

In order to assess how much the Inner Flower

would affect HRV while employed as an explicit brea-

thing guide, further in the experiment we switched-on

the device in the Focus groups and participants car-

ried out the breathing exercise. In the Ambient groups

participants performed a substitute task instead; they

read the short story “The Oval Portrait” by Edgar Al-

lan Poe. Both tasks lasted 6 minutes and were de-

signed to equally involve users’ attention. At the end

of the task, the device was switched-off in the Focus

groups.

Then, all participants performed a second N-back

task (N-back2). Their performance, during this test

would enable us to assess the efﬁcacy of the Inner

Flower as a tool to reduce psychological stress and

improve cognitive availability.

Finally, participants ﬁlled out a third STAI (STAI-

3). Additionally, in the Focus groups, participants

answered the USE questionnaire.

3 RESULTS

Due to the nature of the data and the modest number

of participants per group, we used non-parametric sta-

tistical tests to analyze the data. For studying HRV,

N-back and STAI, we performed resampling (per-

mutation) statistics using the Minque

package from

R. Answers to the USE questionnaire were analyzed

with a Wilcoxon rank sum test. When applicable, we

tested for the effect of each main factor (Attention,

Biofeedback and moment of measurement) as well as

for the interaction of thereof.

3.1 HRV

HRV was computed during both N-back tasks (time1

and time3) and during the breathing exercise (time2).

There was a signiﬁcant interaction between the At-

tention factor and time, with a difference between

Focus during the breathing task (M=1.47×10

−2

SD=0.67×10

−2

) and the rest of the conditions

(M=1.07×10

−2

, SD=0.48×10

−2

, p < 0.05, Figure 6).

Across other factors and interactions there were no

signiﬁcant differences in HRV.

Note that due to technical issues the HR data is

incomplete for 5 participants (out of 36).

3.2 N-back

We investigated the evolution of the percentage of

accuracy during the N-back between the ﬁrst and

https://cran.r-project.org/web/packages/minque/

minque.pdf

Figure 6: HRV across Attention factor and time (i.e. during

N-back1, exercise, N-back2). “*”: p-value < 0.05.

Figure 7: Evolution of the performance between the ﬁrst

and the second N-back task. Performance increased further

in the Focus groups. “**”: p-value < 0.01.

the second test (i.e. N-back2 - N-back1). Over-

all performance increased between those two tests;

with a signiﬁcant effect of the Attention factor. Per-

formance increased more sharply in Focus groups

(M=+11.23pp, SD= 8.36) as compared to Ambient

groups (M=+4.21pp, SD= 7.03, p<0.01, Figure 7).

Across other factors and interactions there was no sig-

niﬁcant differences in the evolution of N-back perfor-

mance.

3.3 STAI

There was a signiﬁcant effect of time. STAI-2 sco-

res (after the ﬁrst N-back task) were higher (M=42.4,

SD=9.08, p < 0.01) when compared to the other sco-

res (M=34.8, SD=8.57, Figure 8, top). There was

a signiﬁcant interaction between Attention and time,

with lower scores in the Focus groups in STAI-3 at

the end of the experiment (M=30.8, SD=5.92), when

compared to the rest of the conditions. (M, 38.5,

SD=9.48, p < 0.01, Figure 8, bottom). Across other

PhyCS 2018 - 5th International Conference on Physiological Computing Systems

Figure 8: Top: STAI scores across time (lower score: less

anxiety). Bottom: STAI scores across Attention factor and

time. “**”: p-value < 0.01.

factors and interactions there was no signiﬁcant diffe-

rences.

3.4 USE Questionnaire

As mentioned earlier the USE questionnaires were

fulﬁlled only in Focus groups. There was no signi-

ﬁcant difference between Static (M=32.8, SD= 2.92)

and Dynamic (M=28.9, SD=6.79, p=0.26) groups.

4 DISCUSSION

The N-back task had a signiﬁcant effect on the percei-

ved anxiety, as the STAI-2 was higher than the STAI-

1. This replicates the validity of such task to induce

stress.

As opposed to previous work – e.g. (Yu et al.,

2015), which employed arithmetic tasks between bre-

athing exercises – we did not measure a noticeable ef-

fect of the Ambient condition and could not validate

H1a. There was no signiﬁcant difference within any

marker of stress (HRV, N-back performance, STAI)

between participants with an active ambient device

and participants with no device (beginning of the

study). We would need longer experiments, or even

longitudinal studies, to better grasp the inﬂuence and

the dynamic of an ambient (and subtle) biofeedback.

Interestingly, when asked after the experiment du-

ring informal interviews, participants in the Ambient

groups reported that they did not pay attention to the

device – an indicator that at the very least it was not

disruptive. Still, it is possible that such an ambient bi-

ofeedback might reduce stress even more than a short

explicit breathing exercise when users are exposed for

a couple hours, or over the course of several days.

In contrast, over the 6-minute breathing exercise

participants of the Focus group demonstrated a lower

level of stress on all three markers – H1b is validated.

Not only did the breathing exercise increased HRV

and helped participants to feel less anxious, but it also

induced a higher performance during the N-back task.

This scenario mirrors the (negative) effects that might

arise when a psychological test is administrated to a

sensitive population in a stressful environment (e.g.

elderly in a hospital). In similar situations, the use of a

device alike the Inner Flower might increase patients’

cognitive availability before a test and prevent biases

during a psychological evaluation.

Since there was no effect of the Feedback moda-

lity over the course of the experiment, we cannot vali-

date H2. An actual biofeedback did not perform better

(or worse) than a simple breathing exercise based on a

ﬁxed breathing rate. There are however multiple fac-

tors that could explain this outcome and that would

need further investigations.

First, by trying to give more freedom to users and

ﬁnely adapt the breathing guide (i.e. calibration phase

in Focus-Dynamic group) we might have introduced

a higher between-subject variability, as we observed

more variability in the Focus-Dynamic group.

Second, the N-back task might have resulted in a

ceiling effect with some participants. While overall

participants tended to improve their scores between

N-back1 and N-back2, those who already had a good

score (i.e. >80%, n=17) had difﬁculties to improve af-

terward. This plateau is among the confounding fac-

tors that we would need to control more ﬁnely du-

ring recruitment for future work, alongside persona-

lity traits.

Third, and maybe most importantly, our choice of

biofeedback might not have been perceived as an ac-

tual manifestation of user’s physiology. While we did

not want to overwhelm users with too many stimuli,

the absence of a dedicated HR biofeedback might

have impeded their sense of agency, i.e. users did not

realize that they were actually “connected” with the

Exploring Biofeedback with a Tangible Interface Designed for Relaxation

device.

This latter interpretation would be on par with the

fact that there was no signiﬁcant difference in the usa-

bility questionnaire. With current results we cannot

validate H3, since Focus-Dynamic and Focus-Static

groups rated equally high the device, between 80%

and 90%. We would expect a higher sense of agency

(i.e. in Focus-Dynamic) to be reﬂected on “Ease of

use” and “Ease of learning” subscales of USE.

Despite the encouraging effect on relaxation and

cognitive availability of the Focus groups, when users

are presented with an explicit breathing guide, the ab-

sence of results between a coupled (Dynamic) and

a ﬁxed (Static) breathing guidance highlights once

again how much rigor is needed when assessing the

effect of physiological measures and the resulting be-

neﬁt of biofeedback.

5 LIMITATIONS AND FUTURE

WORK

As shown by (Hazlewood et al., 2011) it is difﬁcult to

evaluate ambient technology on a one-time basis be-

cause it has to be by deﬁnition blended into the envi-

ronment. In the present study, conditions were maybe

not ecological enough in the sense that experiments

took place in a small and impersonal room. To solve

these issues we are working with a designer to bet-

ter script the overall interaction and bring the study to

a home studio replica. Our next experimental design

will include as well a control group without any de-

vice, for a better control of the confounding variables.

While it is harder to bring the technology outside the

laboratory, longitudinal studies, over several days or

weeks, would inform us about the required amount of

time for an ambient biofeedback to become effective.

Concerning the form factor, we conducted infor-

mal interviews at the end of the experiment. Some of

the participants showed interest in multi-modal feed-

back, for example an audio feedback to let them conti-

nue the breathing exercise while closing their eyes. In

order to enable such usage we will complement ex-

isting feedback, for example with audio waves as in

(Dijk and Weffers, 2011).

To determine the importance of physiological me-

asurements, we hypothesize that a social usage of the

device might be a way to reinforce the association

with an actual biofeedback. (Roseway et al., 2015)

designed an ambient device embedded in the work-

place that informs those around about one’s state.

Thanks to that colleagues were more sensitive to emo-

tional states of others. Similar examples inspired us

to create a collaborative scenario where users would

try to synchronize their heart rate through the device,

with a color-based HR feedback. We envision such

exercise as a way to establish trust between people.

The Inner Flower could then become both a tool to

manage stress and a proxy for communication, for ex-

ample in situations involving care takers and care gi-

vers, or when people suffer from communicative dis-

orders.

Over the course of the study we demonstrated how

a tangible device could help reduce perceived anxiety

(measured with STAI) and alleviate a physiological

symptom of stress (increase in HRV). Moreover, it

enabled participants to improve performance in a cog-

nitive task. Increasing cognitive availability before a

psychological test is one of the most promising ap-

plications of this technology. With a usability score

between 80% and 90% we can state that the device

was appealing to users. In the future we will further

investigate the inﬂuence of both attention and type of

biofeedback.

REFERENCES

Azevedo, R. T., Bennett, N., Bilicki, A., Hooper, J., Mar-

kopoulou, F., and Tsakiris, M. (2017). The calming

effect of a new wearable device during the anticipa-

tion of public speech. Scientiﬁc Reports, 7(1):1–7.

DeBoer, R. W., Karemaker, J. M., and Strackee, J. (1987).

Hemodynamic ﬂuctuations and baroreﬂex sensitivity

in humans: a beat-to-beat model. American Jour-

nal of Physiology-Heart and Circulatory Physiology,

253(3):H680–H689.

Dijk, E. O. and Weffers, A. (2011). Breathe with the ocean:

a system for relaxation using audio, haptic and visual

stimuli. In Special Symposium at EuroHaptics 2010:

Haptic and Audio-Visual Stimuli: Enhancing Expe-

riences and Interaction, Amsterdam, The Netherlands.

Univ. Twente.

Farb, N., Daubenmier, J., Price, C. J., Gard, T., Kerr, C.,

Dunn, B. D., Klein, A. C., Paulus, M. P., and Mehling,

W. E. (2015). Interoception, contemplative practice,

and health. Frontiers in Psychology, 6(June):763.

Fleck, S., Baraudon, C., Frey, J., Lainé, T., and Hachet, M.

(2018). “Teegi’s so Cute!”: Assessing the Pedagogi-

cal Potential of an Interactive Tangible Interface for

Schoolchildren. In IDC 2018 - 17th Interaction De-

sign and Children Conference, Trondheim, Norway.

Frey, J., Grabli, M., Slyper, R., and Cauchard, J. R.

(2018). Breeze: Sharing Biofeedback Through We-

arable Technologies. In CHI ’18.

Hazlewood, W. R., Stolterman, E., and Connelly, K. (2011).

Issues in evaluating ambient displays in the wild: two

case studies. In Proceedings of the SIGCHI confe-

rence on Human factors in computing systems, pages

877–886. ACM.

Health, T., The, S. A., and Hatch, J. P. (1982). Controlled

Group Designs in Biofeedback Research: Ask, "What

PhyCS 2018 - 5th International Conference on Physiological Computing Systems

Does the Control Group Control For?". Biofeedback

and Self-regulation, 7(3).

Kirschbaum, C., Pirke, K.-M., and Hellhammer, D. H.

(1993). The "Trier Social Stress Test" - A Tool for

Investigating Psychobiological Stress Responses in a

Laboratory Setting. Neuropsychobiology, 28(1-2):76–

81.

Kivimäki, M., Leino-Arjas, P., Luukkonen, R., Riihimäi,

H., Vahtera, J., and Kirjonen, J. (2002). Work

stress and risk of cardiovascular mortality: pro-

spective cohort study of industrial employees. BMJ,

325(7369):857.

Lund, A. M. (2001). Measuring usability with the use que-

stionnaire12. Usability interface, 8(2):3–6.

MacLean, K. E. (2009). Putting haptics into the ambience.

IEEE Transactions on Haptics, 2(3):123–135.

Marteau, T. M. and Bekker, H. (1992). The development

of a six-item short-form of the state scale of the Spiel-

berger State-Trait Anxiety Inventory (STAI). British

Journal of Clinical Psychology, 31(3):301–306.

McCraty, R., Atkinson, M., Tomasino, D., and Bradley,

R. T. (2009). The coherent heart: Heart-brain in-

teractions, psychophysiological coherence, and the

emergence of system-wide order. Integral Review,

5(2):10–115.

McKee, M. G. (2008). Biofeedback: An overview in the

context of heart-brain medicine. Cleveland Clinic

Journal of Medicine, 75(SUPPL.2):31–34.

Mehler, B., Reimer, B., and Wang, Y. (2011). A comparison

of heart rate and heart rate variability indices in dis-

tinguishing single task driving and driving under se-

condary cognitive workload. Proceedings of the Sixth

International Driving Symposium on Human Factors

in Driver Assessment, Training and Vehicle Design,

pages 590–597.

Mühl, C., Jeunet, C., and Lotte, F. (2014). Eeg-based wor-

kload estimation across affective contexts. Frontiers

in Neuroscience, 8:114.

Moraveji, N., Olson, B., Nguyen, T., Saadat, M., Khalighi,

Y., Pea, R., and Heer, J. (2011). Peripheral Paced

Respiration: Inﬂuencing User Physiology during In-

formation Work. the 24th annual ACM symposium

on User interface software and technolog, pages 423–

428.

Muench, F. (2008). HRV: The manufacturers and vendors

speak. Association for Applied Psychophysiology &

Biofeedback, 36(1):35–39.

Owen, A. M., McMillan, K. M., Laird, A. R., and Bullmore,

E. (2005). N-back working memory paradigm: A

meta-analysis of normative functional neuroimaging

studies. Human Brain Mapping, 25(1):46–59.

Roseway, A., Lutchyn, Y., Johns, P., Mynatt, E., and Czer-

winski, M. (2015). BioCrystal : An Ambient Tool for

Emotion and Communication. International Journal

of Mobile Human Computer Interaction, 7(3):20–41.

Schein, M., Gavish, B., Herz, M., Rosner-Kahana, D., Na-

veh, P., Knishkowy, B., Zlotnikov, E., Ben-Zvi, N.,

and Melmed, R. (2001). Treating hypertension with a

device that slows and regularises breathing: a rando-

mised, double-blind controlled study. Journal of Hu-

man Hypertension, 15(4):271–278.

Spielberger, C. D., Gorsuch, R. L., and Lushene, R. E.

(1970). Manual for the state-trait anxiety inventory.

Palo Alto, CA: consulting psychologists press.

Tan, G., Dao, T. K., Farmer, L., Sutherland, R. J., and Ge-

virtz, R. (2011). Heart rate variability (hrv) and post-

traumatic stress disorder (ptsd): A pilot study. Applied

Psychophysiology and Biofeedback, 36(1):27–35.

Vaschillo, E. G., Vaschillo, B., and Lehrer, P. M. (2006).

Characteristics of Resonance in Heart Rate Variability

Stimulated by Biofeedback. Applied Psychophysio-

logy and Biofeedback, 31(2).

Yu, B., Feijs, L., Funk, M., and Hu, J. (2015). Breathe

with touch: a tactile interface for breathing assistance

system. In Human-Computer Interaction, pages 45–

52. Springer.

Yucha, C. B. and Montgomery, D. (2008). Evidence-based

practice in biofeedback and neurofeedback. AAPB

Wheat Ridge, CO.

Exploring Biofeedback with a Tangible Interface Designed for Relaxation