ITACA Neurofeedback Training Tutorial

Welcome to the ITACA Neurofeedback training app tutorial! Here we will explain step by step how to configure the different options presented by ITACA for the design and development of Neurofeedback training. 

Getting started: ITACA configuration panel

ITACA is designed to be a flexible framework to facilitate the design of different Neurofeedback protocols. This means that its configuration panel presents multiple options (see Fig. 1). The different configuration options are divided into the following sections: (i) General settings; (ii) Neurofeedback settings; (iii) Signal processing settings; (iv) Time settings; and (v) Scenario settings. Below, we will go through each of the following sections specifying each of the parameters to be configured.  In addition, the panel has 4 main buttons: 

• Load: Allows to load a JSON file with the training configuration. This is intended to facilitate the reproducibility of the experiments. 
• Reset: Sets the default configuration. 
• Save: Allows to save the current configuration as a JSON file. 
• Done: Closes the configuration panel and confirms the selected options. If there was a problem (i.e., an inconsistent definition of the training frequency bands or an important parameter not selected), a dialogue will be displayed, preventing the panel from closing. 

General settings

In this section you can find the fields User, Session, and Run (see Fig. 1). These fields allow you to store useful information for identifying and managing the different recordings throughout a Neurofeedback study.

Scenario field displays the different Neurofeedback training scenarios provided by ITACA: Mental cube, Trapped spaceship and Neuro runner. For more information on these training scenarios, see Functionalities of this app. 

NF training mode allows you to select one of the 5 Neurofeedback training modes (i.e., metrics) implemented in ITACA: 

• Power band training: The feedback signal will be derived from the power of the training frequency band measured over the chosen channels. This metric requires at least one channel to be selected.
• Power ratio bands training: The feedback signal will be extracted from the ratio between de power of two frequency bands measured over the chosen channels. This metric requires at least one channel to be selected.
• Global coupling training: The feedback signal will be derived from the average of all connectivity values from adjacency matrix. This metric selects all channels automatically.
• Node strength training: The feedback signal will be extracted as the strength of the selected channel (or average of strengths if more than one channel are selected). This metric requires at least one channel to be selected.
• Node coupling training: The feedback signal will be extracted as the connectivity values from adjacency matrix between the channels selected (or average of connectivity values if more than two channels are selected). This metric requires at least two channels to be selected.

Connectivity metric field allows to select the functional connectivity metric used to calculate the adjacency matrix in NF training modes based on functional connectivity. The metrics provided are weighted phase lag index (wPLI) and orthogonalized amplitude envelope correlation (AEC-ort). The pipelines of both metrics have been carefully desgined and optimized to ensure a correct functioning of real-time feedback presentation. This field is only enable when a training mode based on functional connectivity is selected. 

Channel selection  has two fields: First, on the left, a drop-down menu opens to select the standard used in the assembly (i.e., 10-20, 10-10 or 10-05). On the other hand, on the right, the Selection button displays an interface that shows the channels used in the recording (see Fig. 2).  This interface allows you to select the channels from which you want to calculate the training metric. Furthermore, it is possible to save  the channel configuration as JSON file and load it using Save and Load buttons respectively. The selected channels will be displayed in the main GUI in the text box Training channels.

Neurofeedback settings

This section contains the options related to the training frequency band as well as the rejection of noisy signal segments (see Fig. 3). First, the Training band selection table displays one row (or two rows if Power ratio band training is selected) to define the lower and upper limits (in Herzts) of the training band.  Additional frequency bands can be defined in the Power ratio band training mode. For instance, the denominator of the ratio can be defined to consist of the power of the theta (4-8 Hz) and beta (13-20 Hz) bands. To do this, press the Add band button (only active in the Power ratio band training mode). A new row will then appear to define the additional band. In the first column, select whether the band is to be part of the numerator or denominator of the ratio. 

Below this table you can see two "radio" buttons: Increase activity and Decrease activity. These options are mutually exclusive.  While the Increase activity option calculates the feedback signal proportionally to the brain activity analysis metric (i.e., an increase of the metric corresponds to an increase of the feedback signal), the Decrease activity option will show an increase of the feedback signal when the value of the trained brain activity decreases. This option is useful when training is aimed at reducing abnormally increased brain rhythms (e.g., beta activity in the basal ganglia in Parkinson's patients). 

Finally, there are the options for rejecting noisy signal segments. These options allows to analyze if the time window used to calculate the feedback signal in the online mode has certain characteristics to be classified as noisy or containing artifacts. Therefore, these tools seek to avoid the influence of components not associated with brain activity on the feedback signal. Specifically, two frequency bands can be selected that have typically been associated with two types of artifacts: Ocular noise (0.5 - 2 Hz), and Muscular noise (> 35 Hz). When either of these two options is selected, the statistical properties of the signal in its associated frequency bands during the baseline period are calculated. The allowed percentage increase values (spin boxes to the right of each option) define thresholds of variation of its characteristics above which the signal is considered too noisy or contains artifacts. In this case, the returned feedback signal is None, avoiding the influence of these noisy periods.

Signal processing settings

This section contains the options related to the time parameters taken during the real-time calculation of the feedback signal from the EEG. The Feature window field allows the user to select the width of the EEG window used to calculate the chosen analysis metric. This time window is taken from the most recent EEG signal samples at each time point.

Update rate field displays the rate at with the training metric is calculated (in milliseconds). Lastly, Baseline recording field allows to browse and load a baseline recording from MEDUSA© Recorder app to extract the baseline value. Other types of recordings are not supported.  This baseline value is used as reference to correct the metric values calculated during closed-loop training. If no reference record is loaded, there will be a calibration period at the start of the application. See the Time settings section for more on this.

Lastly, two check boxes allow to select what spatial filters to apply in real-time signal processing. On the one hand, the common average reference (CAR). It can be chosen for both power-based and functional connectivity-based training modes. This filter subtracts from each channel the common mean values of all channels, so it is only useful when we have a large and homogeneous set of EEG channels in the whole head. Otherwise, its use is not recommended. On the other hand, the surfaces laplacian filter.  It can be only selected for power-based training modes. This is an automated version of the surface Laplacian filter that determines the best channels surrounding the target channels to enhance their local activity and reject common activity. Its use is recommended only when a set of EEG channels homogeneously distributed around the training channels is available. Otherwise, the application of the laplacian surface filter will not be optimal. 

Time settings 

In this section the different options allow you to configure the time structure of the Neurofeedback training (see Fig. 5). First of all, it is necessary to define some concepts of the temporal organization of the ITACA sessions. Each time ITACA's Neurofeedback training app is launched (i.e., configured and "powered" in MEDUSA© Platform), it is considered a different "run". It is important to note that only one training scenario can be used in each run. Each "run" is organized in different "trials". All of them will share the same configuration in terms of scenario objective. All trials will have a maximum duration, after which, if the user has not reached the scenario objective (e.g., modulate his brain activity above a certain value), the trial will end. If the scenario objective is reached earlier, the trial will be terminated. When all trials have been completed, the "run" is considered to be over. When a run is completed, a dialog is displayed to save the EEG signal as well as other data associated with the recording. For more information, see the section Runing a training scenario at the end of this tutorial. Through this temporal structuring, ITACA allows to define different "runs", varying the training scenario or the difficulty of the target scenario, in the same Neurofeedback session.

Number of trials fields allows defining how many "trials" will played in the incoming run. Time between trials field allows setting the resting time between each "trial" of the same "run". Maximum trial duration sets the duration of the "trial" if the user does not reach the scenario objective. Calibration time field is only needed if a baseline recording is not loaded. As mentioned above, calibration time should be long to extract enough time windows for the calculation of the baseline value.

Scenario settings 

In this section different settings will be displayed depending on the training scenario selected under General settings. For more information on training scenarios, see Functionalities of this app. 

Mental cube settings

Mental cube scenario has two parameters to adjust (see Fig. 6). Threshold sets a percentage value with respect to the baseline value so that if the value of the metric is below the threshold, no feedback of that value will be given. For example, with a 5% as Threshold, all values of the training metric that are not higher than the baseline value by at least 5% will be ignored. Since in this scenario the cube will increase its vertical position by an amount proportional to the training metric at each point in time, this threshold prevents the goal from being reached by small random changes. 

The Speed parameter controls the steps of increasing/decreasing the height of the cube at any given time. In this sense, it is worth noting that the height update is calculated as follows:

h_new = h_current + x_feedback * speed,

where h_new is the updated value, h_current the previous height of the cube, and x_feedback  the feedback value calculated as the difference between the metric value in that time point and the baseline value. Therefore, a higher Speed value will lead to greater jumps in cube's height. This parameter can be adjusted to reach the target when it is difficult for the user to reach it. The default parameter value ensures a smooth height update under normal conditions. 

Trapped spaceship settings

Trapped spaceship scenario allows to set the parameter that determines the difficulty in reaching the target (see Fig. 7). The target is defined as a maximum increase of the trained feature with respect to the baseline value (in percentage). In this way, the maximum height to which the ship has to be brought on the stage is always the same, but will be matched with different incremental values depending on the defined target. This makes it possible to increase the difficulty of Neurofeedback training by modifying this parameter. The height of the spaceship at each time point is calculated as follows: 

h_new = ((h_up - h_low)/ Δpct_max) * ((x_feedback - x_baseline)/|x_baseline|) + h_low ,

where h_up and h_low are the maximum and minimum heights the spaceship can reach, respectively,  Δpct_max is the maximum increase of the trained feature expected to reach, and x_feedback and  x_baseline are the metric values in that time point and in the the baseline period, respectively. Therefore, the first factor in the equation determines a parameter to map the expected maximum feature increase with the maximum height the spacecraft will reach, and the second factor calculates the feedback value as the change in the percentage of the training feature with respect to its reference value. 

Neuro runner settings

Neuro runner scenario has 4 parameters to adjust, 2 of which are used to adjust the difficulty of the scenario (see Fig. 8) . On the one hand, as in Trapped spaceship scenario, the Target (expressed as a maximum increase in percentage of training feature  with respect to the baseline value) at which the runner controlled by the user reaches the maximum speed. On the other hand, the speed of the Opponent NPC. This is also expressed in terms of increase with respect to the baseline value. In this regard, it should be noted that the Opponent NPC will move at a constant speed bounded by this parameter. For example, taking the values shown in Fig. 8: The user-controlled avatar will move at maximum speed when its activity reaches a 20% increase from the baseline, while the Opponent NPC will move at a constant speed corresponding to half the maximum speed the user could reach. Therefore, the difficulty of this scenario can be increased either by increasing the Target, by increasing the speed of the Opponent NPC, or by combining both parameters.  Of note, the calculation followed to map the training feature value and the runner speed is the same as the followed in  Trapped spaceship: 

s_new = ((s_max)/ Δpct_max) * ((x_feedback - x_baseline)/|x_baseline|),

where s_max is the maximum speed that the runners can reach,  Δpct_max is the maximum increase of the trained feature expected to reach, and x_feedback and  x_baseline are the metric values in that time point and in the the baseline period.

In addition, this scenario allows a choice of avatars, both the user-controlled and the NPC Opponent. These are chosen in the Runner and Opponent fields. Among the options are two male and two female athletes, a robot, a dummy, an alien and an astronaut. In addition, the avatars will perform a celebratory dance when they reach the finish line! 

Running a training scenario 

Do you already have your training experiment set up? Then let's see how to properly run an NF training scenario. 

First of all, it is recommended to save the training configuration using the Save button in the configuration panel. This will allow you not to waste time in setting up the whole experiment in future training sessions. 

To start the application we press the Power button in MEDUSA Platform with the ITACA Neurofeedback Training app selected. As we can see, after a few moments, a message will appear in the console: "Setting up TCP server...". Although we see that after pressing the Power button, the button to its right changes to show a Play button (green arrow), this should not be pressed yet. Shortly after the above message appears in the console, our app window will open with the following message: "Waiting to start". At that moment another message will appear in the MEDUSA Platform console: "Unity is ready to start". This indicates that we can now press the Play button to start the run. 

Once the training run has started, a message will appear in the center of the screen: "Starting session". This allows the user to prepare for the start. Then, if a file with a recording for the calculation of the baseline value has not been loaded, the calibration will take place. During this period the message "Calibrating" will appear on the screen and the user should remain calm, avoiding blinking or moving to avoid introducing artifacts in the baseline signal.

After the calibration time has elapsed, the first Neurofeedback training trial will start.  Our training scenario will show the time remaining to complete the trial in the upper left corner, and the trial number in which we are in the lower right corner. 

When the training target is reached, the following message will appear "Well done!", otherwise "Trial finished" will appear. There will then be a rest time between trials, which can be configured in the configuration panel. After the rest time has elapsed, a message with a countdown will warn the user that a new trial is about to start. 

Once all trials have been completed, the message "Session finished" will appear. At this point, you can return to the MEDUSA Platform and press the Stop button (red square). This will close the app window and then a dialog will appear to save the signal recorded during the Neurofeedback session (see Fig. 9). As can be seen, the user id, the session and run numbers are placed in file information. By default, the file will be named after the date of registration. By clicking on the three dots you can specify the name and where to save the file. 

In addition to the signal, a series of data will be stored in the file that may be of interest in a later analysis of the recording. These are when was the onset of each trial, how long each trial lasted, if the target was reached or not in each trial, if and when the trial was paused, when the trial was restarted after pausing it, the feedback values calculated, the timestamps in which these feedback values were calculated and the baseline value. These data are included in the class NeurofeedbackData. 

For the moment, this is all that is needed to set up and conduct Nuerofeedback training using the ITACA Neurofeedback Training app. We hope you enjoy it and find it useful. If you have any problem, doubt or suggestion, do not hesitate to write in our forum.