Multi-Modal Integrated Mini-QEEG Solution with Results, Training

Protocols and Neurofeedback in Real-time

Francisco Marques-Teixeira

, Horácio Tomé-Marques

, João Andrade

and João Marques-Teixeira

1,2

Neurobica, Neurobios - Instituto de Neurociências, Rua Agostinho de Campos, 369, 4200-015, Porto, Portugal

Laboratory of Neuropsychophysiology, University of Porto, Rua Alfredo Allen, 4200-135, Porto, Portugal

Keywords: Neurofeedback, BCI, Brain-Computer_Interaction, EEG, Electroencephalography, QEEG, Quantitative EEG,

Mini-QEEG, Dysfunctional Patterns, EEG Phenotypes.

Abstract: Many Neurofeedback softwares allow clinicians to develop modular protocols. Some even allow to be per-

formed quantitative Electroencephalography (QEEG) analysis. However, an all-in-one solution does not exist,

with built-in decision tree, that permits to perform a QEEG analysis, apply a protocol decision, and perform

subsequent iterated mini-QEEG analysis, and subsequente protocol decisions and training, in the same sys-

tem. Our application, besides being able to make the identification of the brain dysfunctional patterns con-

cerning its electrophysiology, and accurately choose the Neurofeedback training protocols to apply, performs

real-time Neurofeedback. We compared our system with a conventional EEG and QEEG system in a proof of

concept rational, obtaining an average Pearson Correlation Coefficient of 0.89 regarding the dysfunctional

patterns and protocols, as well as remitted dysfunctional patterns. In conclusion, our application is pervasive,

scalable and potentially ubiquitous. And it can be extended into multiple and different consumer fields.

1 INTRODUCTION

This extended abstract aims to present the model be-

hind an application for Neurofeedback (NF) training.

Our model purposes a multi-modal integrated mini

quantitative Electroencephalography (mini-QEEG)

solution with results, training protocols and NF in

real-time. Firstly, we present the theory behind brain

performance optimization and how it differs from

clinical NF. Secondly, we present how this theory can

be applied to our software architecture model and

how the system architecture is organized. To finalize,

we present some preliminary results on how this sys-

tem is as accurate and valid as a conventional clinical

EEG and NF system.

Neurofeedback is a brain performance technique

that uses processes of operant conditioning which leads

to self-regulation of brain activity (Ute Strehl, 2014).

Recent studies suggest that EEG Neurofeedback repre-

sents feasible and promising a tool for therapeutic in-

terventions, and cognitive enhancement (Enriquez-

Geppert et al, 2017), and that critical brain dynamics

can be modulated with closed-loop stimulation in an

automatic, involuntary fashion (Zhigalov et al, 2016).

We developed an application that processes EEG

signal from 16 channels in real-time, computes met-

rics, e.g., power, frequency, phase, for each electrode,

and, according to inter- and intra-hemispheric physi-

ological ratios and asymmetries of the main fre-

quency bands (theta, alpha and beta), it assesses the

main dysfunctional brain electrophysio-logical pat-

terns in each user, and the main Neurofeedback pro-

tocols to perform according to a severity order.

We also programmed a decisional tree that as-

sesses the success of each Neurofeedback session and

outputs which protocol of the list should be done af-

terwards, up to 10 sessions per protocol. A user’s

block is composed of 3 protocols and, in the end of

the block, another QEEG is automatically registered

for verification of the overall improvement. If the im-

provement is satisfactory according to a defined

threshold, the training is finished. However, if not,

another block of 1 Neurofeedback protocol (up to 10

sessions) is recommended. The program repeats this

process until the improvements are satisfactory.

2 METHODS

To develop this application, we have used the Pro-

cessing Software, based in Java, to write and compile

all the code.

The system architecture was based on a fixed

main steps model:

Marques-Teixeira F., TomÃl’-Marques H., Andrade J. and Marques-Teixeira J.

Multi-Modal Integrated Mini-QEEG Solution with Results, Training Protocols and Neurofeedback in Real-time.

In NEUROTECHNIX 2017 - Extended Abstracts (NEUROTECHNIX 2017), pages 5-9

1. Pre-processing: EEG signal Processing;

2. Post-processing: metrics computation;

3. QEEG: rules and protocols decisional tree;

4. Neurofeedback: mini-QEEG and training

analysis

5. Validation of the whole application and ex-

tracted metrics.

For the first (1) step, we applied Fast Fourier

Transform (FFT), with a 0.98 Hz resolution, to 16

scalp electrodes (Fp1, Fp2, Fz, F3, F4, Cz, C3, C4,

T3, T4, Pz, P3, P4, O1, O2, Oz (this channel is repre-

sented as the average between O1 and O2). We use

linked-ears reference and ground is located on AFz

site. Based on the frequency spectrum, a known lead-

off detection current and Ohm’s Law, we also com-

pute the impedance in each electrode.

For the second (2) process, we computed the spec-

tral power for each electrode and for different fre-

quency bins and frequency-bands, in real-time.

We also developed a blink and jaw clench detec-

tion routine. When one of these events is detected, the

software assumes the value of the referred metrics as

the result of the linear interpolation of the last 4 val-

ues computed.

For the third (3) process of this work, we com-

puted dysfunctional patterns, such as dysfunctional

power ratios, asymmetries inter-hemispheric and in-

tra-hemispheric, among others.

These dysfunctional patterns were based in devi-

ations of the normal brain electrophysiology, grouped

as instabilities (frontal alpha asymmetry, frontal beta

asymmetry, antero-posterior beta inversion) (Da-

vidson, 1979, 2004; Harmon-Jones, 2004; Minnix et

al., 2004; Nitschke, et al., 2004; Sutton and Davidson,

1997; Wiedemann et al., 1999), disconnection (dis-

rupted Hibeta at T3 and/or T4) (Coan and Allen,

2004), blocking (disrupted Hibet:beta ratio at Fz

and/or Pz, also known as Swingle ratio) (Swingle and

Paul, 2015); hot temporals (high percentage of beta

and hibeta at T3 and/or T4), and dystonus (hypertonus

and hypotonus) (Hagemann, 2004).

Our software has a table of rules of dysfunctional

patterns grouped according to the groups referred

above and every-time there is a ratio that is inverted

or not-present the rule is turned-on, thus, the dysfunc-

tional pattern is considered.

Our software ranks the dysfunctional patterns of

the user according to an serial order, based on their

score, and link that dysfunctional order to a specific

training protocol. As soon as a NF training session is

completed, the efficiency of the protocol is automati-

cally assessed by a mini-QEEG, in such a way that

the program will decide if the user will need to train

the same protocol again or will jump to the next one

in the serial order. In this application, the minimum

training sessions per protocol are 6 sessions and the

maximum are 10 sessions. Each training block is

composed by 3 training protocols and each block has

a minimum of 18 sessions (3x6) and a maximum of

30 sessions (3x10).

In the end of each block, the program performs a

post-block QEEG to assess the success of the training.

If the overall dysfunctional patterns are not corrected,

the client will be indicated to do another block, this

process being repeated until the dysfunctional pat-

terns normalize. All the computation is done instanta-

neously.

For the fourth (4) process, we have programmed a

real-time feedback related to a base-line measured in

each training session that indicates the positive and

negative feedback to be given to the user according to

the dysfunctional pattern rule (reinforce or inhibit a

certain frequency band in specific electrodes). This

feedback is adaptive relative to a moving base-line, of

60 seconds, that dictates its difficulty level. Thus, we

created a generative feedback, according to the level

of the user’s learning.

Figure 1: Example of the generative adaptation of the feed-

back of one user. First plot – user’s performance in blue and

feedback given in orange. Second plot – Maximum limit

used to compute the feedback. Third plot – feedback given

(from worst to best: red, pink, dark blue, light blue). We can

see that the maximum limit goes up because the metric be-

ing trained rises in the moving base-line. Thus, the user

stops receiving a positive feedback because now the level

is more difficult. Then, since it became “too hard”, the max-

imum limit drops down again, and the user starts receiving

more positive feedback again.

To validate (5) this application:

1) we compared it to a standard QEEG system;

2) and perform NF sessions to assess the training

efficiency of the system.

Regarding the QEEG system comparison we used

the Neuronic E8.5 system (neuronicsa.com) for the

EEG registration and the Neuroguide software (ap-

pliedneuroscience.com) for editing and processing

EEG data.

We have made 4 recordings with 4 different users,

3 males and 1 female, mean age 34,75, with eyes

closed. Each recording is composed by 2 consecutive

1 minute acquisitions using Neurobica (our system)

and Neuronic. The same electro-cap (http://electro-

cap.com) was used with both systems and only the

connections to the EEG acquisition system was

changed, so all the practical physical conditions were

maintained for both systems. We verified the imped-

ances with Neurobica, maintaining their values below

10 KΩ before recording the signal.

Regarding the clinical validation of the NF train-

ing, 1 user made 11 sessions of approximately 5

minutes, in two days (6 in the first and 5 in the fol-

lowing day). All sessions were done with the same

NF protocol (based on the reinforcement of β band in

F3 channels relatively to F4) to study its metric evo-

lution with the training.

3 RESULTS

In this section, we present the preliminary results for

the comparison of the metrics (a) computed with the

Neuronic system and ours (b) as well as the Neu-

rofeedback training results.

3.1 Systems Comparison

Figure 3 shows the comparison between the results of

the 2 sets of one minute acquisitions, Neuronic and

Neurobica, for the four users included in the study.

This comparison was done based on the values com-

puted for the metrics corresponding to the dysfunc-

tional patterns scores. Neuronic values were com-

puted offline, after acquisition, using MATLAB

(www.mathworks.com), and Neurobica results were

computed online by our application.

Figure 2: Comparison between results computed after of-

fline processing of Neuronic acquisition (x axis) and online

processing during Neurobica acquisition (y axis).

Figure 3: Dysfunctional pattern metric score across NF

training sessions in the first (above) and second (below)

days. Some values are missing because they were not rec-

orded. Optimal value should be above 1. In the beginning

of the first day, the score is significantly below that value.

It rises as the sessions are made, until it reaches a value

higher than 1. In the second day, it maintains its value

around 1.

As it can be seen in the plots of the correlation results

are statistically significant (for all users, p-value <

0.001), with an average Pearson Correlation Coeffi-

cient of 0.89.

3.2 Neurofeedback Training

For preliminary assessment of NF training efficiency,

one users spent two consecutive afternoons perform-

ing NF with the same protocol. Each NF session

lasted for 5 minutes and small breaks between ses-

sions were made. The specific dysfunctional pattern

metric relative to the protocol was computed before

and after NF training. Figure 3 shows this metric evo-

lution across time in the two afternoons of training.

4 DISCUSSION

4.1 Systems Comparison

The preliminary results indicate that the online anal-

ysis performed by the Neurobica system is strongly

correlated with the offline analysis made after the ac-

quisition with Neuronic, which leads us to suggest

that we are measuring and computing reliable data.

4.2 Neurofeedback Training

Neurofeedback training preliminary results are very

satisfying and encouraging because they show the ef-

fectiveness of the NF training in the regularization of

the dysfunctional pattern metric score, in the first day.

They also show that, after regularized, the metric sta-

bilizes in the following day. This is an important con-

clusion relatively to Neurofeedback effectiveness in

general and more studies based on this approach are

going on, with more users and different follow-up

periods.

4.3 Future Work

For this work, we programmed all the metrics and ra-

tios computation. Efforts are being made for in the fu-

ture to work on the mining of all the intra and inter

user data in a local server. During this process, all the

data will be returned to the users in a dashboard with

the following information: 1) scheme of the training

sessions accomplished according with the training

proposed by the application that self-organizes ac-

cording to the success in the correction of the dys-

functional patterns; 2) information about the training

protocol used in each session and how was the evolu-

tion of the metrics and rations of that protocol; 3) in-

formation about the dysfunctional patterns assesses in

the QEEG of each user; 4) topographic maps of the

respective metrics and ratios of each user; 5) we are

also working on the assessment of cognitive perfor-

mance improving, symptom reduction and behav-

ioural and well-beingness changes after the

Neurofeedback training and ; 6) we are preparing the

extension this pilot study to more subjects in order to

accomplish the final validation of this integrated

system.

ACKNOWLEDGEMENTS

This project is funded by the program Horizon 2020.

REFERENCES

Davidson, R.. J., Schwartz, G. E., Saron, C., Bennett, J.,

Goleman, D. J., 1979. Frontal versus parietal EEG

asymmetry during positive and negative affect. Psycho-

physiology 16, 202–203.

Davidson, R. J., 2004. What does the prefrontal cortex

“do” in affect: Perspectives on frontal EEG asymmetry

research. Biological Psychology 67, 219–233.

Coan, J. A. Allen, J. J. B., 2004. EEG Asymmetry as a

moderator and mediator of emotion. Biological Psy-

chology 67, 7–49.

Enriquez-Geppert, S., Huster, R. J., and Herrmann, C. S.

2017. EEG-Neurofeedback as a Tool to Modulate Cog-

nition and Behavior: A Review Tutorial. Frontiers in

Human Neuroscience, 11, 51.

Hagemann, D., 2004. Individual differences in anterior

EEG-asymmetry: Methodological problems and solu-

tions. Biological Psychology 67, 157–182.

Harmon-Jones, E., 2004. Contributions from research on

anger and cognitive dissonance to understanding the

motivational functions of asymmetrical frontal brain

activity. Biological Psychology 67, 51–76.

Minnix, J. A., Kline, J. P., Blackhart, G. C., Pettit, J. W.,

Perez, M., Joiner, T. E., 2004. Relative left frontal

activity is associated with increased depression in high

reassurance-seekers. Biological Psychology 67,

145–155.

Nitschke, J. B., Heller, W., Etienne, M. A., Miller, G. A.,

2004. Prefrontal cortex activity differentiates processes

affecting memory in depression. Biological Psychology

67, 125–143.

Sutton, S. K., Davidson, R. J., 1997. Prefrontal brain asym-

metry: A biological substrate of the behavioural ap-

proach and inhibition systems. Psychological Science

8, 204–210.

Wiedemann, G., Pauli, P., Dengier, W., Lutzenberger, W.,

Birbaumer, N., Buchkremer, G., 1999. Frontal brain

asymmetry as a biological substrate of emotions in pa-

tients with panic disorders. Archives of General Psy-

chiatry 56, 78–84.

Strehl, U. 2014. What learning theories can teach us in de-

signing neurofeedback treatments. Frontiers in Human

Neurosciences, 8: 894.

Swingle and Paul, G. 2015. Adding Neurotherapy to Your

Practice.

Zhigalov A., Kaplan A., Palva J. M. 2016. Modulation of

critical brain dynamics using closed-loop neurofeed-

back stimulation. Clinical Neurophysiology, Volume

127, Issue 8 , 2882 – 2889.