Towards Bidirectional Brain-computer Interfaces that Use

fNIRS and tDCS

Samuel W. Hincks

, Maya DeBellis

, Eun Youb Lee

, Ronna ten Brink

, Birger Moëll

and Robert J. K. Jacob

Tufts University, Medford, MA 02155, U.S.A.

Stockholm University, 11418 Stockholm, Sweden

Keywords:

BCI, Implicit Brain-computer Interface, fNIRS, Near-infrared Spectroscopy, tDCS, Transcranial Direct

Current Stimulation, Cognitive Workload, Bidirectional Brain-computer Interface, Entropic Brain-computer

Interface, N-back, ADHD, Attention.

Abstract:

We envision a future user interface that measures its user’s mental state and responds not only through a

display but also by sending output directly to the brain, leading to a primitive bidirectional brain-computer

interface. Previous interactive systems have measured brain state with functional near-infrared spectroscopy

(fNIRS) for communication from user to computer; we now explore transcranial direct-current stimulation

(tDCS) as a channel in the opposite direction. Our goal is to integrate this with brain measurements from

fNIRS, so that the stimulation parameters governing tDCS may be set dynamically to enhance user cognition

based on current mental state and task demands. To do this, the ﬁrst step is to determine how long it takes for

tDCS to register cognitive effects and how long these effects last. We present an experiment that investigates

the temporal dimension of tDCS for this purpose. The ﬁndings suggest a long lag-time between the onset

of stimulation and any measurable cognitive effect, which may prohibit the effectiveness of tDCS in a brain-

adaptive application.

1 INTRODUCTION

Computers support several methods for communi-

cating with the user, but currently these output

methods are constrained by users’ sensory channels.

Non-invasive brain stimulation techniques, such as

transcranial direct-current stimulation (tDCS), might

transcend this limitation. Evidence in the psychology

literature suggests that tDCS can temporarily enhance

or emphasize aspects of user cognition (Brunoni and

Vanderhasselt, 2014) without imposing a health risk

(Bikson et al., 2016). tDCS delivers a weak (1 to 2

milliamp) electrical current to the exterior of the sub-

ject’s scalp through an electrode, taking the path to

the nearest cathode, which has been carefully placed

so that the current will enter and alter particular re-

gions of the subject’s brain. tDCS has been used to

treat depression (Nitsche et al., 2009), as well as en-

hance language learning (Flöel et al., 2008), working

memory (Fregni et al., 2005) and attention (Gladwin

et al., 2012). With the introduction of tDCS to the

standard output arsenal of HCI, an interactive system

may be able to judiciously enhance these abilities de-

pending on the circumstance and user state.

Our study is aimed at a future user interface that

uses brain measurement as input and responds not

only with the usual screen output but also by send-

ing output directly to the brain, suggesting a primitive

bidirectional brain-computer interface. Previous sys-

tems have measured brain state with fNIRS for com-

munication from user to computer (Afergan et al.,

2014a; Afergan et al., 2015; Solovey et al., 2012);

we now explore tDCS for the opposite direction in a

bidirectional brain-computer interface, with fNIRS or

another brain monitor as input, and tDCS as output.

For example, consider a brain-adaptive UAV system

(Afergan et al., 2014a). In this experiment, we trained

machine learning algorithms operating on data from

an fNIRS brain monitor to predict the user’s cogni-

tive workload as he or she was controlling the ﬂight

paths of several simulated unmanned aerial vehicles

(UAV). The system added extra UAVs to the user’s

task when brain activity indicated that she was in a

state of low cognitive workload; and it removed some

in order to simplify the user’s task when workload in-

creased. The bidirectional version we propose would

Hincks, S., DeBellis, M., Lee, E., ten Brink, R., Moëll, B. and Jacob, R.

Towards Bidirectional Brain-computer Interfaces that Use fNIRS and tDCS.

DOI: 10.5220/0006380500570064

In Proceedings of the 4th International Conference on Physiological Computing Systems (PhyCS 2017), pages 57-64

ISBN: 978-989-758-268-4

apply tDCS stimulation brieﬂy, precisely when the

measured workload increases and only for the du-

ration of the workload spike. For such a bidirec-

tional brain-computer interface to work in practice,

the lag time between stimulation and its result should

be short. However, much previous tDCS research,

especially experiments aimed at treating depression

(Shiozawa et al., 2014), have emphasized longer term

effects and longer stimulation periods, because inter-

activity was not the goal.

To proceed with an interactive system, the key

question is to determine how long it takes for tDCS

stimulation to register cognitive effects and how long

these effects last. We investigate that in this paper

with two experiments. At present, there are many un-

knowns regarding the relationship between settings of

the device and its associated cognitive effects, mak-

ing it difﬁcult to gauge whether the device warrants

study and inclusion in next generation user inter-

faces. An alarming percentage of experiments (Hor-

vath et al., 2015) fail to elicit signiﬁcant improve-

ments to user performance. The consensus is that re-

sults vary across person, possibly because each indi-

vidual has a different brain and unique rules for how

to conduct brain stimulation. However, deciding to

abandon tDCS for that reason is premature, because

the device has not yet been studied interactively. The

missing ingredient for effective tDCS may in fact be

a two-way digital ecosystem in which settings can be

dynamically adjusted based on their judged, subject-

to-subject effectiveness.

In this paper, we evaluate the feasibility of a bidi-

rectional brain-computer Interface. We present two

experiments aimed at estimating temporal properties

of tDCS by estimating performance changes in a visu-

ospatial n-back task over a 15 minute time-course. In

the ﬁrst experiment, we compare 5 minutes of tDCS

stimulation to a placebo condition; and in the second

experiment, we compare 10 minutes of stimulation to

a placebo condition. We evaluate changes in reaction

time and accuracy for each minute of the experiment.

2 BACKGROUND

2.1 Transcranial Direct Current

Stimulation

While introducing tDCS brain stimulation into HCI

raises safety and ethical questions, research to date

has shown that when stimulation does not exceed 2

milliamps and lasts shorter than 40 minutes, there

have been no cases of irreversible injury caused by

tDCS in a sample of 33,200 sessions (Bikson et al.,

2016). Compared to other brain stimulation tech-

niques, tDCS is easy to use and potentially inexpen-

sive; it already supports a do-it-yourself community

(Fitz and Reiner, 2013). Although experiments typi-

cally use a more advanced setup, the basic device con-

sists of just two electrodes and a battery to energize

them. Direct current is then administered through a

saline-soaked sponge or rubber electrode with con-

ductive gel on the subject’s scalp. In a typical setup

(and the one used in this experiment), one electrode is

placed over the target of stimulation, initiating a path

for the current to take to a second electrode placed

somewhere nearby. The current is presumed to al-

ter the cortical excitability of the neurons it interacts

with, either depolarizing their membranes and mak-

ing the neurons more likely to ﬁre in the case of an-

odal stimulation, or hyperpolarizing the membranes,

making the neurons less likely to ﬁre in the case of

cathodal stimulation (Medeiros et al., 2012).

Given that working memory, compulsivity, and

attention are often impaired in individuals with

Attention-deﬁcit/hyperactivity disorder (ADHD)

(Moëll et al., 2015), tDCS has been explored as treat-

ment for individuals with this condition (Cachoeira

et al., 2017). Experiments aimed to enhance working

memory typically administer anodal stimulation to

the left dorsolateral prefrontal cortex (dlPFC) at the

site F3 (in the International 10-20 system (Nitsche

et al., 2008)) and allow current to ﬂow through a

reference electrode at a symmetrical location on the

brain’s right hemisphere at site F4 (Zaehle et al.,

2011) (see Figure 1). Many experiments have used

this montage to enhance performance at an n-back

test (Brunoni and Vanderhasselt, 2014). The present

experiment makes use of the same montage and

n-back paradigm, except we investigate shorter

stimulation periods and track performance on a

minute-by-minute basis in order to evaluate the usage

of tDCS in an interactive system.

2.2 Functional Near Infrared

Spectroscopy

fNIRS is a non-invasive neuroimaging technique,

which can be implemented cheaply since it consists

merely of light sources and detectors (Piper et al.,

2014). fNIRS depicts brain activation by shining

near-infrared light into the scalp and detecting the

amount that returns to the sensor, which changes

based on the relative concentration of oxygenated and

deoxygenated hemoglobin, the basic energy supply

for neurons. These measurements have been found to

ﬂuctuate in response to the user’s cognitive workload

(Herff et al., 2014).

PhyCS 2017 - 4th International Conference on Physiological Computing Systems

Figure 1: The n-back task, pictures of the device, and one

subject’s fNIRS activity.

2.3 N-Back

Cognitive workload is a basic index of the cerebral

strain a task poses on a user (Herff et al., 2014). We

focus on the workload in the prefrontal cortex, which

generally correlates with short-term memory work-

load. In experiments, this type of workload is typi-

cally induced using an n-back task. In the visuospa-

tial variety of this task (see Figure 1), the user tracks

a 3-by-3 grid where one box is colored. The colored

box changes every couple seconds, and the user’s job

is to indicate whether or not the colored box is in the

same location as it was n iterations ago. Higher val-

ues of n induce higher degrees of short term memory

workload. tDCS has been reported to improve n-back

performance (Brunoni and Vanderhasselt, 2014) and

fNIRS has been reported to differentiate brain sig-

nals pertaining to trials with higher or lower associ-

ated cognitive workload (Herff et al., 2014; Hincks

et al., 2016).

2.4 Implicit Brain-computer Interfaces

Implicit Brain-Computer Interfaces (Zander et al.,

2014; Treacy Solovey et al., 2015) listen to run-

ning classiﬁcations of the user’s state as measured

by portable brain sensors (such as EEG and fNIRS),

and update implicit system settings to the user’s cur-

rent needs. Using fNIRS, predictions of the user’s

cognitive workload have been applied to control cur-

sor selection expansion (Afergan et al., 2014b), mu-

sical scores (Yuksel et al., 2016), robot automation

(Solovey et al., 2012), and task difﬁculty (Afergan

et al., 2014a). Likewise, this measure could drive the

administration of tDCS.

3 EXPERIMENTS

3.1 Equipment

For measuring brain activity, we used the multichan-

nel frequency domain Imagent fNIRS device from ISS

Inc. (Champaign, IL) to acquire brain data. It uses

two probes, each with four light sources emitting light

at 830 and 690 nanometers, and detectors located

between 0.5 and 3.5 centimeters away from these

sources. Sampling frequency was set to 11.79hz.

For altering brain activity, we used Soterix 4x1

HD-tDCS multi-channel stimulation interface (model

4X1-C2) to pass electrical currents and the Soterix

tDCS-CT (model 1507-LTE) to control stimulation

and placebo according to a double-blind protocol.

3.2 Experiment 1

Nine undergraduate college students (5 female) par-

ticipated in the ﬁrst experiment. They were mone-

tarily compensated and gave consent at the beginning

of the experiment. A university Institutional Review

Board approved the experiment. The experimenter

explained the visual n-back task (Figure 1) on a white-

board, and let the user practice two trials of the 1-

back and two trials of the 2-back. For the 1-back, the

user hit the left arrow key if the visual arrangement

matched the previous one and the right arrow key oth-

erwise, and for the 2-back they indicated whether or

not it matched what they saw 2 iterations ago. These

keys were marked with ‘YES’ and ‘NO’ with tape

on the keyboard. This task was implemented with

custom software for the purpose of recording reac-

tion time and dynamically labeling fNIRS data. Af-

ter these practice trials, the experimenter ﬁt the user

with tDCS and fNIRS. This entailed ﬁrst measuring

the size of the subject’s head and selecting between

four cap sizes, and then placing one gel-covered an-

odal electrode at site F3 and the other reference elec-

trode at site F4 (Nitsche et al., 2008), and then con-

necting the electrodes to the Soterix device (see Fig-

ure 1). Next, we placed the two fNIRS probes as near

as possible to those sites. (We do not report on any

fNIRS data in this paper for experiment one or two

because we were unable to discover stimulation de-

pendent patterns).

The subsequent experiment proceeded in two

phases. In the ﬁrst phase, subjects alternated between

30 seconds of the 1-back and 30 seconds of the 2-

back, performing each task 7 times. This served as

practice as well as the opportunity to group partici-

pants by the separability of their fNIRS data. In the

second phase, the subject alternated between 40 sec-

Towards Bidirectional Brain-computer Interfaces that Use fNIRS and tDCS

Table 1: Differences in N-Back Accuracy and Reaction Time for each Minute of Experiment 1.

Percent Accuracy Reaction Time (milliseconds)

Sham Real Sham Real

min mean std dev mean std dev p-value mean std dev mean std dev p-value

1 74 24 77 22 0.82909 930 247 808 209 0.4575

2 82 34 93 10 0.58809 701 216 852 250 0.36142

3 80 39 90 14 0.65056 770 261 880 41 0.43568

4 81 28 98 5 0.27549 808 310 811 40 0.98716

5 84 29 88 10 0.84721 651 189 832 166 0.17768

6 82 34 93 15 0.60227 744 201 728 167 0.90031

7 78 32 100 0 0.2236 681 224 691 96 0.9418

8 83 28 93 10 0.52828 669 198 718 149 0.69088

9 80 33 87 13 0.70177 631 218 699 146 0.61433

10 82 29 98 5 0.33798 655 230 651 73 0.97642

11 84 35 95 10 0.57964 637 182 723 109 0.43732

12 80 39 89 16 0.67354 658 159 763 93 0.28151

13 81 28 90 8 0.54059 666 273 620 81 0.75824

14 84 35 90 8 0.76657 590 168 704 149 0.32322

15 78 26 92 5 0.34072 624 201 715 187 0.50996

m 81 31 91 4 0.53000 694 209 746 52 0.65

onds of the 2-back and 20 seconds of rest, repeating

this 15 times for a total of ﬁfteen minutes. In the n-

back task (for experiment 1 and 2), a new stimulus

appeared every 3 seconds, and accuracy and reaction

time for the 40 second task was therefore based on the

average of 13 responses.

We used a between subject design. Prior to the

experiment, the participant had been placed in two

groups: four in the real tDCS group and ﬁve in the

sham group, and neither experimenter nor subject

knew the groups. The real group received 2 milliamps

of anodal stimulation at site F3 for 5 minutes. The

sham group received 2 milliamps of stimulation only

for 30 seconds, a standard placebo, since subjects tend

to sense when the device turns on but forget about it

when it has been on for a while (Brunoni and Vander-

hasselt, 2014). Participants began the experiment in

parallel to onset stimulation. Afterwards, the exper-

imenters removed the equipment from the user and

debriefed them.

Results: We have summarized the results of the

ﬁrst experiment in Table 1, and there were no sig-

niﬁcant effects for the 5 minute stimulation, although

stimulated user’s trended towards better accuracy and

the control group trended towards faster speed, hint-

ing more at a speed-accuracy trade-off than cognitive

enhancement. Table 1 shows the mean and standard

deviation of the participants’ mean accuracy and re-

action time for each of the ﬁfteen trials under both

sham and real conditions, as well as the probability

that these averages differed between sham and real

conditions in an independent t-test. Without a clear

indication that 5 minutes of stimulation exerted sig-

niﬁcant improvements to user performance, we mod-

iﬁed our design and conducted a second experiment.

3.3 Experiment 2

Fourteen college students (4 female) participated in

the second experiment. Based on the lack of sig-

niﬁcant results in the ﬁrst experiment, we increased

stimulation time from 5 to 10 minutes, and used a

within subject design so that all participants received

both real and sham stimulations. Participants alter-

nated whether or not they received real stimulation

ﬁrst, and both experimenter and subject were blind to

this information. To allow time for both conditions,

we removed the initial ﬁfteen minute practice period,

and participants alternated between 1-backs and 2-

backs, starting with the 1-back. For both real and

sham stimulation, participants thus completed 8 sets

of 40-second 1-back and 8 sets of 2-backs with a 20

second rest in between. In total, each condition lasted

sixteen minutes, separated by a ﬁve minute break. Be-

cause interference from hair prevented fNIRS mea-

surement in the ﬁrst experiment, we placed the two

fNIRS probes on the user’s forehead. Apart from

these changes, the second experiment proceeded iden-

tically to the ﬁrst.

Results: We have summarized the results of the

second experiment in Table 2, which is arranged iden-

tically to Table 1, and illustrate changes in accuracy

in Figure 2 and changes in reaction time in Figure

3. Overall, tDCS did not signiﬁcantly improve ei-

ther n-back accuracy or reaction time after 10 minutes

of stimulation. However, there was a signiﬁcant im-

PhyCS 2017 - 4th International Conference on Physiological Computing Systems

Table 2: Differences in N-Back Accuracy and Reaction Time for each Minute of Experiment 2.

Percent Accuracy Reaction Time (milliseconds)

Sham Real Sham Real

min mean std dev mean std dev p-value mean std dev mean std dev p-value

1 93 7 95 6 0.3028 716 167 719 219 0.9216

2 95 7 95 5 0.7184 851 260 799 203 0.2297

3 94 11 92 9 0.7229 799 193 706 222 0.0629

4 95 8 94 7 0.7001 770 260 783 341 0.8458

5 93 12 94 11 0.5162 731 208 669 205 0.1777

6 91 10 88 15 0.4098 793 249 843 355 0.7569

7 93 10 95 8 0.4657 684 157 684 154 0.7898

8 90 13 95 7 0.1775 751 278 709 217 0.2580

9 88 9 97 6 0.0034** 741 204 639 185 0.0323*

10 94 9 91 10 0.2456 749 230 771 313 0.9678

11 90 9 95 8 0.0631 720 177 628 234 0.0975

12 99 5 95 8 0.2519 762 299 707 277 0.2690

13 93 7 91 12 0.7966 649 157 626 193 0.5743

14 96 7 89 20 0.1930 701 211 715 270 0.9637

15 88 12 89 9 0.7260 624 144 595 155 0.2914

16 94 11 98 4 0.1944 640 198 710 314 0.4852

m 93 7 93 5 0.5049 746 185 706 223 0.3618

provement to n-back accuracy during the last minute

of stimulation. For minute 9-10, the mean accuracy

of the 1-back in the sham condition was 88% (std =

9) and the mean accuracy in the 10 minute stimula-

tion condition was 97% (s = 6) (N =13, p = 0.0034

in a paired sample t-test). At minute 9-10, improved

accuracy in the 1-back did not come at the expense of

speed. In fact, reaction times in the 10 minute stimu-

lation condition (m = 639 ms, s = 185 ms) were sig-

niﬁcantly faster than reaction times in the sham con-

dition (m = 741 ms, std = 204 ms) (N =13, p = 0.0323

in a paired sample t-test).

Note that since 1 out of 20 tests should be signif-

icant with a threshold set to 0.05, it is hard to verify

whether variation has occurred due to chance or not.

If signiﬁcance thresholds are modiﬁed according to a

Bonferroni correction, then the new threshold is 0.05

/ 16 = 0.003125 since there are 16 tests, and neither

accuracy nor reaction time are signiﬁcantly better in

the stimulation condition than in the sham condition,

although accuracy at minute 9 misses Bonferonni cor-

rected signiﬁcance by less than 0.0003. There are two

reasons why the results between minute nine and ten

could be regarded as more valid. First, signiﬁcance

occurs at the very last minute of stimulation and not in

a more random minute during the ten stimulation min-

utes or ﬁve non-stimulation minutes. Second, the two

dependent variables exhibiting a statistically signif-

icant effect according to non-conservative statistical

thresholds refer to the same minute, which is improb-

able unless there was a true effect driving enhance-

ment at this minute, especially given the expectation

of a speed accuracy trade-off.

Figure 2: Changes in percent accuracy over time, recorded

at the end of each minute.

Figure 3: Changes in reaction time measured in millisec-

onds, recorded at the end of each minute.

Towards Bidirectional Brain-computer Interfaces that Use fNIRS and tDCS

4 DISCUSSION

According to these results, tDCS requires at least 9

minutes of stimulation in order to register an effect.

Whether or not effects escalate beyond 10 minutes

is an interesting investigation for future work. For

present purposes, the delayed response between stim-

ulation and effect implies that fNIRS-adaptive stim-

ulation using tDCS may not work effectively. In

the interactive application that motivated the design

of this experiment, a subject would perform a com-

puter task under the interrogation of fNIRS measure-

ment, and tDCS would apply stimulation to the user

when brain activation measures indicated that cogni-

tive workload had increased. The results indicate that

the user would need to wait at least 9 minutes before

enjoying a boost to cognition, and a brain-adaptive

deployment of the technology would therefore be ap-

plicable to tasks with a time span in this range. This

is feasible in practice, but less amenable to study in

an experimental setting.

It is not clear why it takes 9 minutes of stimula-

tion for behavioral effects to register nor whether this

limitation disappears given better settings to the de-

vice. Individual differences in skin texture, bone den-

sity, and brain structure may imply that standardized

stimulation protocols fail to appropriately customize

to any given subject. If that is the case, better set-

tings to device parameters such as intensity, polarity,

duration, and probe location could be discovered and

change based on simultaneous brain measurements

(McKendrick et al., 2015).

We envision a design in which fNIRS could mon-

itor the relative activation of the user’s task-positive

and task-negative networks, which oscillate in inverse

correlation to each other depending on whether or not

the user is sensorily immersed or in an introspective

mode of cognition (Raichle et al., 2001). The back-

and-forth activity of these networks could be moni-

tored; a bidirectional brain-computer interface might

discover how to stimulate the user’s brain in order

to maximize task-positive immersion and minimize

task-negative introspection when warranted. A ﬁrst

step in this direction would be to evaluate whether

or not fNIRS detects short term neurobiological re-

actions to stimulation. We attempted such an investi-

gation in this experiment, but in the ﬁrst experiment

hair prevented our device from appropriately measur-

ing the targeted F3 and F3 nodes. We note that other

fNIRS devices (such as Hitachi ETG 4000) can solve

this problem. In the second experiment, when probes

were placed approximately 3 inches from the site of

stimulation, we did not observe any obvious fNIRS

patterns separating the stimulation and real condi-

tions. However, we found no severe limitations pre-

venting the two devices from being used in concert.

Our target is an interactive system in which real-time

fNIRS measurements are used to modify the tDCS

stimulation parameters for better effectiveness. Our

experimental conﬁgurations and results present a ﬁrst

step in support of such a bidirectional brain-computer

interface.

5 FUTURE WORK: ENTROPIC

BRAIN-COMPUTER

INTERFACING

Because of the lag-time between the onset of stimu-

lation and any measurable cognitive effect, research

into bidirectional brain-computer interfacing might

instead focus on stimulation modalities with a more

immediate impact on the user’s mental state. Re-

search suggests that listening to music with lyrics is

detrimental to performance on tasks that require con-

centration (Shih et al., 2012), but the conclusion is

unclear for music without lyrics, which may or may

not enhance cognition depending on task, user, and

song. As with tDCS, there likely exists some cog-

nitively enhancing stimulation procedure for a given

user although the exact procedure may vary person-

to-person and the necessary information to determine

which stimulation procedure to administer is encoded

as a physical constellation in the user’s brain.

The problems of cognitive enhancement via music

and electrical stimulation may have similar computa-

tional solutions, and should thus be studied in con-

cert. In both cases, variables controlling procedures

that output physical events to the brain (as current

or sound) need to be conﬁgured such that stimula-

tion is both safe and beneﬁcial to the recipient. For

tDCS, these variables describe the location of electri-

cal probes applying current of a given intensity and

polarity for a duration of time. For music, the vari-

ables of interest govern a procedure for generating an

array of decibel amplitudes in the frequency domain.

The collective work of music theory describes

rules for producing harmonious sound, restricting the

large space of sound possibilities. For example, a

given sequence of sounds should be organized around

some tonic note (or frequency series) known as the

key of the song, and harmonious sounds are math-

ematically related to this fundamental frequency ac-

cording to some scale of intervals (e.g., the major or

minor scales). Engineers have encoded these rules

into digital audio workstations, enabling an opportu-

nity for auditory bidirectional brain-computer in-

PhyCS 2017 - 4th International Conference on Physiological Computing Systems

terfacing if the song is augmented with an interface

that allows it to branch between different versions de-

pending on implicit input from sensors measuring the

user’s physiology. For this research, we recommend

music production via the Web Audio API (Rogers,

2012) and open source javascript software that ex-

tends it (Choi and Berger, 2013; Mann, 2015) since

adaptive songs written using web tools can be played

in a browser and distributed online for use by anyone

with access to the Internet.

In (Hincks et al., 2017), we refer to information-

based bidirectional brain-computer interfacing as en-

tropic brain-computer interfacing to emphasize a use-

ful model of consciousness that relates subjective

measures of rich experience to fMRI-based measures

of system entropy (quantiﬁed as the difﬁculty to pre-

dict future states of the brain from previous states)

(Carhart-Harris et al., 2014). This model and the

larger enterprise of Bayesian cognitive neuroscience

suggests that a major goal of the brain is to optimally

compress and learn from sensory data. The brain

uses existing models to predict the content of sen-

sory signals, and propagates information which vio-

lates expectation up cognitive hierarchies where ex-

isting models are modiﬁed (Friston, 2010). Music -

or sound which obeys mathematical patterns - may

exist as a happy coincidence of the brain’s procliv-

ity to direct computation (and associated conscious

experience) towards stimuli which engages its predic-

tive machinery (Huron, 2006). By this reasoning, the

state of the brain and attention can be modulated by

manipulating the user-relative predictability of sound.

Several components of a system which adapts

sound to physiological measures of the brain could

be recycled to perform bidirectional brain-computer

interfacing in other modalities (e.g. electrical stim-

ulation). This generic brain optimization algorithm

hinges on non-invasive methods for detecting some

aspect of user cognition worth optimizing in all cases.

In (Hincks et al., 2017), we argue for two user dimen-

sions describing the direction (internal vs. external

origin) and intensity (high vs. low entropy) of atten-

tion and attempt to measure these states using fNIRS

and EEG. If these states are measured in real-time and

allowed jurisdiction over variables governing concur-

rent stimulation, a machine learning algorithm could

infer a relationship between state transitions and the

variables governing stimulation, so that the system

over time learned how to coerce desirable states.

ACKNOWLEDGEMENTS

We thank Tad Brunye, Erika Hussey, Leanne Hirsh-

ﬁeld, Tomoki Shibata, Daniel Afergan, Beste Yuksel,

Remco Chang, Evan Peck, Angelo Sassaroli, and Ser-

gio Fantini who are students, alumnni, and professors

at Tufts University.

Research was sponsored by the U.S. Army Nat-

ick Soldier Research, Development and Engineer-

ing Center,and was accomplished under Cooperative

Agreement Number W911QY-15-2-0001. The views

and conclusions contained in this document are those

of the authors and should not be interpreted as rep-

resenting the ofﬁcial policies, either expressed or im-

plied, of the U.S. Army Natick Soldier Research, De-

velopment and Engineering Center, or the U.S. Gov-

ernment. The U.S. Government is authorized to re-

produce and distribute reprints for Government pur-

poses notwithstanding any copyright notation hereon.

REFERENCES

Afergan, D., Hincks, S. W., Shibata, T., and Jacob, R. J.

(2015). Phylter: a system for modulating notiﬁcations

in wearables using physiological sensing. In Inter-

national Conference on Augmented Cognition, pages

167–177. Springer.

Afergan, D., Peck, E. M., Solovey, E. T., Jenkins, A.,

Hincks, S. W., Brown, E. T., Chang, R., and Jacob,

R. J. (2014a). Dynamic difﬁculty using brain metrics

of workload. In Proceedings of the 32nd annual ACM

conference on Human factors in computing systems,

pages 3797–3806. ACM.

Afergan, D., Shibata, T., Hincks, S. W., Peck, E. M., Yuksel,

B. F., Chang, R., and Jacob, R. J. (2014b). Brain-

based target expansion. In Proceedings of the 27th

annual ACM symposium on User interface software

and technology, pages 583–593. ACM.

Bikson, M., Grossman, P., Thomas, C., Zannou, A. L.,

Jiang, J., Adnan, T., Mourdoukoutas, A. P., Kronberg,

G., Truong, D., Boggio, P., et al. (2016). Safety of

transcranial direct current stimulation: evidence based

update 2016. Brain Stimulation, 9(5):641–661.

Brunoni, A. R. and Vanderhasselt, M.-A. (2014). Working

memory improvement with non-invasive brain stimu-

lation of the dorsolateral prefrontal cortex: a system-

atic review and meta-analysis. Brain and cognition,

86:1–9.

Cachoeira, C. T., Leffa, D. T., Mittelstadt, S. D., Mendes,

L. S. T., Brunoni, A. R., Pinto, J. V., Blazius,

V., Machado, V., Bau, C. H. D., Rohde, L. A.,

et al. (2017). Positive effects of transcranial direct

current stimulation in adult patients with attention-

deﬁcit/hyperactivity disorder–a pilot randomized con-

trolled study. Psychiatry Research, 247:28–32.

Carhart-Harris, R. L., Leech, R., Hellyer, P. J., Shanahan,

M., Feilding, A., Tagliazucchi, E., Chialvo, D. R., and

Towards Bidirectional Brain-computer Interfaces that Use fNIRS and tDCS

Nutt, D. (2014). The entropic brain: a theory of con-

scious states informed by neuroimaging research with

psychedelic drugs.

Choi, H. and Berger, J. (2013). Waax: Web audio api ex-

tension. In NIME, page 499–502.

Fitz, N. S. and Reiner, P. B. (2013). The challenge of craft-

ing policy for do-it-yourself brain stimulation. Jour-

nal of medical ethics, pages medethics–2013.

Flöel, A., Rösser, N., Michka, O., Knecht, S., and Breit-

enstein, C. (2008). Noninvasive brain stimulation im-

proves language learning. Journal of Cognitive Neu-

roscience, 20(8):1415–1422.

Fregni, F., Boggio, P. S., Nitsche, M., Bermpohl, F., An-

tal, A., Feredoes, E., Marcolin, M. A., Rigonatti, S. P.,

Silva, M. T., Paulus, W., et al. (2005). Anodal tran-

scranial direct current stimulation of prefrontal cortex

enhances working memory. Experimental brain re-

search, 166(1):23–30.

Friston, K. (2010). The free-energy principle: a uni-

ﬁed brain theory? Nature Reviews Neuroscience,

11(2):127–138.

Gladwin, T. E., den Uyl, T. E., Fregni, F. F., and Wiers,

R. W. (2012). Enhancement of selective attention by

tdcs: interaction with interference in a sternberg task.

Neuroscience letters, 512(1):33–37.

Herff, C., Heger, D., Fortmann, O., Hennrich, J., Putze, F.,

and Schultz, T. (2014). Mental workload during n-

back task—quantiﬁed in the prefrontal cortex using

fnirs. Frontiers in human neuroscience, 7:935.

Hincks, S. W., Afergan, D., and Jacob, R. J. (2016). Us-

ing fnirs for real-time cognitive workload assessment.

In International Conference on Augmented Cognition,

pages 198–208. Springer.

Hincks, S. W., Bratt, S., Poudel, S., Phoha, V., Dennett,

D. C., Jacob, R. J. K., and Hirshﬁeld, L. M. (2017).

Entropic brain-computer interfacing: Using fnirs and

eeg to measure attentional states in a bayesian frame-

work. In PhyCS.

Horvath, J. C., Forte, J. D., and Carter, O. (2015). Evidence

that transcranial direct current stimulation (tdcs) gen-

erates little-to-no reliable neurophysiologic effect be-

yond mep amplitude modulation in healthy human

subjects: a systematic review. Neuropsychologia,

66:213–236.

Huron, D. B. (2006). Sweet anticipation: Music and the

psychology of expectation. MIT press.

Mann, Y. (2015). Interactive music with tone.js. In Proceed-

ings of the 1st annual Web Audio Conference. Citeseer.

McKendrick, R., Parasuraman, R., and Ayaz, H. (2015).

Wearable functional near infrared spectroscopy (fnirs)

and transcranial direct current stimulation (tdcs):

expanding vistas for neurocognitive augmentation.

Frontiers in systems neuroscience, 9:27.

Medeiros, L. F., de Souza, I. C. C., Vidor, L. P., de Souza,

A., Deitos, A., Volz, M. S., Fregni, F., Caumo, W.,

and Torres, I. L. (2012). Neurobiological effects of

transcranial direct current stimulation: a review.

Moëll, B., Kollberg, L., Nasri, B., Lindefors, N., and

Kaldo, V. (2015). Living smart—a randomized con-

trolled trial of a guided online course teaching adults

with adhd or sub-clinical adhd to use smartphones to

structure their everyday life. Internet Interventions,

2(1):24–31.

Nitsche, M. A., Boggio, P. S., Fregni, F., and Pascual-

Leone, A. (2009). Treatment of depression with tran-

scranial direct current stimulation (tdcs): a review. Ex-

perimental neurology, 219(1):14–19.

Nitsche, M. A., Cohen, L. G., Wassermann, E. M., Priori,

A., Lang, N., Antal, A., Paulus, W., Hummel, F., Bog-

gio, P. S., Fregni, F., et al. (2008). Transcranial direct

current stimulation: state of the art 2008. Brain stim-

ulation, 1(3):206–223.

Piper, S. K., Krueger, A., Koch, S. P., Mehnert, J., Haber-

mehl, C., Steinbrink, J., Obrig, H., and Schmitz, C. H.

(2014). A wearable multi-channel fnirs system for

brain imaging in freely moving subjects. Neuroimage,

85:64–71.

Raichle, M. E., MacLeod, A. M., Snyder, A. Z., Powers,

W. J., Gusnard, D. A., and Shulman, G. L. (2001). A

default mode of brain function. Proceedings of the

National Academy of Sciences, 98(2):676–682.

Rogers, C. (2012). Web audio api. s Draft [online speciﬁ-

cation], Version, 1.

Shih, Y.-N., Huang, R.-H., and Chiang, H.-Y. (2012).

Background music: Effects on attention performance.

Work, 42(4):573–578.

Shiozawa, P., Fregni, F., Benseñor, I. M., Lotufo, P. A.,

Berlim, M. T., Daskalakis, J. Z., Cordeiro, Q., and

Brunoni, A. R. (2014). Transcranial direct current

stimulation for major depression: an updated system-

atic review and meta-analysis. International Journal

of Neuropsychopharmacology, 17(9):1443–1452.

Solovey, E., Schermerhorn, P., Scheutz, M., Sassaroli, A.,

Fantini, S., and Jacob, R. (2012). Brainput: enhanc-

ing interactive systems with streaming fnirs brain in-

put. In Proceedings of the SIGCHI conference on

Human Factors in Computing Systems, pages 2193–

2202. ACM.

Treacy Solovey, E., Afergan, D., Peck, E. M., Hincks, S. W.,

and Jacob, R. J. (2015). Designing implicit interfaces

for physiological computing: guidelines and lessons

learned using fnirs. ACM Transactions on Computer-

Human Interaction (TOCHI), 21(6):35.

Yuksel, B. F., Oleson, K. B., Harrison, L., Peck, E. M.,

Afergan, D., Chang, R., and Jacob, R. J. (2016). Learn

piano with bach: An adaptive learning interface that

adjusts task difﬁculty based on brain state. In Proceed-

ings of the 2016 CHI Conference on Human Factors

in Computing Systems, pages 5372–5384. ACM.

Zaehle, T., Sandmann, P., Thorne, J. D., Jäncke, L., and

Herrmann, C. S. (2011). Transcranial direct current

stimulation of the prefrontal cortex modulates work-

ing memory performance: combined behavioural and

electrophysiological evidence. BMC neuroscience,

12(1):1.

Zander, T. O., Brönstrup, J., Lorenz, R., and Krol, L. R.

(2014). Towards bci-based implicit control in human–

computer interaction. In Advances in Physiological

Computing, pages 67–90. Springer.

PhyCS 2017 - 4th International Conference on Physiological Computing Systems