EEG brain-computer interface Direct Brain-to-Brain Communication in Humans:  
A Pilot Study

August 12, 2013
Home Experiment Researchers Neural Systems Laboratory CCD Laboratory Contact
We sought to demonstrate that it is possible to send information extracted from one brain directly to another brain, allowing the first subject to cause a desired response in the second subject through direct brain-to-brain communication. A task was designed such that the two subjects could cooperatively solve the task by transmitting a meaningful signal from one brain to the other. 

Experimental Set-Up
The experiment leverages two existing technologies: electroencephalography (or EEG) for noninvasively recording brain signals from the scalp and transcranial magnetic stimulation (or TMS) for noninvasively stimulating the brain.

Figure 1 illustrates the experimental paradigm. Electrical brain activity from Subject 1 (the "Sender") is recorded using EEG (Figure 2) in the Neural Systems Laboratory in the Computer Science and Engineering building at UW. This brain activity is interpreted by a computer and is transmitted (when classified as a valid motor imagery signal) over the internet to the TMS machine in the Institute for Learning and Brain Sciences (I-LABS) building at UW. The TMS machine delivers a magnetic stimulation pulse to the left motor brain region of Subject 2 (the "Receiver"), causing the right hand to press a key (Figure 3). 

The Task
The task that the subjects must cooperatively solve via brain-to-brain communication is a computer game (Figure 4). The task involves saving a "city" (on the left) from getting hit by rockets fired by a "pirate ship" from the lower right portion of the screen (depicted by skull-and-bones). To save the city, the subjects must fire a "cannon" located at the lower center portion of the screen. If the "fire" button is pressed before the moving rocket reaches the city, the rocket is destroyed (Figure 5), the city is saved, and the trial ends. To make the task more interesting, on some trials, a friendly "supply plane" may appear instead of a pirate rocket and move leftwards towards the city (Figure 6).  The subjects must avoid firing the cannon at the supply plane.

Brain-to-Brain Collaboration between the Two Subjects
Only Subject 1 (the "Sender") watches the game (Figure 2, Sender watching the game screen, which is not shown). The Sender is unable to press the "fire" button which is only available to Subject 2 (the "Receiver"). The Sender can however engage in motor imagery of the right hand (i.e., imagine moving their right hand) - this imagery signal is recognized by the computer and translated to a magnetic stimulation pulse that is delivered to the left motor cortex region of the Receiver. The stimulation causes a quick upward jerk of the Receiver's right hand, which is resting slighty above the "fire" key on a keyboard (Figure 3). This up-down movement of the hand typically (though not always) results in the "fire" key being hit, causing the cannon in the computer game to be fired as requested by the Sender. If the moving target happens to be the supply plane, the Sender can choose not to fire the cannon at the plane by resting and not engaging in any motor imagery.

Decoding Motor Imagery from EEG
Electrical signals were recorded from the Sender's scalp using the noninvasive technique of EEG.  We used a USBamp EEG recording system (Guger Technologies, Austria) with gold-plated electrodes placed over the left hemisphere at standard locations under the 10-20 convention (sampling frequency = 256Hz). A Laplacian spatial filter was used to reduce artifacts common to nearby electrodes and emphasize local activity. The power in a low frequency band (the "mu" band) was computed across the electrodes and the electrode most correlated with the subject's motor imagery during an initial training period was selected as the control electrode for the task. Changes in the "mu" band have long been linked to motor imagery signals and used in BCIs (for an introduction, see [1]).  The computer translated the power in the mu band to upward movement of a cursor (left side of Figure 4). Specifically, right hand imagery typically causes a decrease in power which was mapped to upward movement of the cursor. If the cursor hit the blue circular target at the top, the computer decides that the Sender has engaged in motor imagery and sends a stimulation pulse to the Receiver.

Stimulation using TMS
The Receiver receives information from the Sender via transient noninvasive brain stimulation induced by a rapidly changing magnetic field. The magnetic field is generated by a special coil held over the Receiver’s head and kept in place by an orientable mechanical arm mounted on the Receiver’s chair. This stimulation technique is called Transcranial Magnetic Stimulation (TMS) and is a well-established noninvasive means to directly influence the activity of a specific spot on the brain’s surface. Note that no electrical pulse is given to the Receiver and stimulation is induced only indirectly through the changing magnetic field. A MagStim Rapid2 model TMS machine was used with a single-pulse TMS protocol. The pulse was delivered though a circle coil at 69% of the machine's power output. The TMS coil was localized over the part of the Receiver’s brain that controls the wrist and fingers. The Receiver kept one or more fingers on a standard computer keyboard’s space bar (the designated "fire" key). The TMS pulse produced a muscle twitch and upward hand movement, typically resulting in the space bar being hit as a result and firing the cannon in the Sender's computer game.

The pilot study involved 2 subjects, both researchers involved in the study (R. Rao as "Sender" and A. Stocco as "Receiver"). The pilot study was approved by the University of Washington Institutional Review Board (IRB) within the Human Subjects Division.

Four experimental sessions were conducted, each with 5-7 trials. Session 1 was terminated early due to network communication issues, which were resolved as the subjects waited for the next session. Sessions 2, 3, and 4 witnessed successful transmission of information from the Sender to the Receiver.

In Sessions 2 and 3, while the Receiver's performance was highly accurate (~90% success rate in stimulation causing the hand to move and the cannon being fired), the Sender's performance was closer to chance levels as the subject reported being in the process of learning to generate the appropriate signal (imagery or rest) given the type of target.

In Session 4, both the Sender and Receiver achieved close to perfect performance (Sender: 100% correct detection and transmission of appropriate signal, Receiver: 100% correct elicitation of hand movement upon stimulation; 1 stimulation not causing the "fire" key to be hit). A portion of the session log is given below ("Missile" = pirate rocket, "Airplane" = supply plane) :
2013-08-12 15:47:37.472000: Starting experiment
2013-08-12 15:47:47.549000: Starting trial: Airplane
2013-08-12 15:47:48.008000: arming stimulator by key press
2013-08-12 15:48:04.106000: Missiles hit: 0, Airplanes hit: 0, Attempts: 1
2013-08-12 15:48:14.165000: Starting trial: Missile
2013-08-12 15:48:17.200000: BCI input received.  Sending TMS pulse.
2013-08-12 15:48:19.691000: Shot fired
2013-08-12 15:48:19.691000: Missiles hit: 1, Airplanes hit: 0, Attempts: 2
2013-08-12 15:48:29.757000: Starting trial: Missile
2013-08-12 15:48:30.062000: BCI input received.  Sending TMS pulse.
2013-08-12 15:48:32.417000: Shot fired
2013-08-12 15:48:32.417000: Missiles hit: 2, Airplanes hit: 0, Attempts: 3
2013-08-12 15:48:42.460000: Starting trial: Missile
2013-08-12 15:48:54.340000: BCI input received.  Sending TMS pulse.
2013-08-12 15:49:03.423000: Missiles hit: 2, Airplanes hit: 0, Attempts: 4
2013-08-12 15:49:13.457000: Starting trial: Airplane
2013-08-12 15:49:29.572000: Missiles hit: 2, Airplanes hit: 0, Attempts: 5
2013-08-12 15:49:39.615000: Starting trial: Missile
2013-08-12 15:49:40.910000: BCI input received.  Sending TMS pulse.

The results suggest that information extracted noninvasively from one brain using EEG can be transmitted to another brain noninvasively using TMS to allow two persons to cooperatively solve a task via direct brain-to-brain transfer of information. The next phase of the study will attempt to quantify this transfer of information using a larger pool of human subjects.

Background and References
  1. For background on brain-computer interfacing, see: 
  2. For a description of related experiments by other groups, see the recent article: 

B2B diagram
Figure 1. Experimental Set-Up. Brain signals from Subject 1 (the "Sender") were recorded using EEG. When imagined hand movements were detected by the computer, a "fire" command was transmitted over the internet to the TMS machine, which caused an upward movement of the right hand of Subject 2 (the "Receiver"), usually resulting in the "fire" key being hit. 

EEG set up
Figure 2. EEG signals being recorded from Subject 1 (the "Sender") as the subject watches the computer game (the game screen is to the left and not shown in the picture). The larger screen displays EEG signals processed by the BCI2000 software. The smaller laptop screen placed further away is from the live Skype session and shows Subject 2 in the TMS lab across campus. Researcher Dev Sarma monitors the experiment. (Picture by researcher Bryan Djunaedi)

TMS set up
Figure 3. Subject 2 (the "Receiver") with TMS coil placed over left motor cortex region and right hand resting slightly above the "fire" key on  the keyboard. The screen behind the subject shows the Sender's game screen which is not seen by the Receiver.

cannon game
Figure 4. Screen shot of the cannon game. A pirate ship on the right side (skull-and-bones) shoots a rocket towards a city on the left. The Sender engages in motor imagery to move the white cursor on the left to hit the blue target.

Target hit
Figure 5. If the Sender is able to use motor imagery to move the white cursor to the circular target (which turns red), a stimulation signal is sent to the Receiver located elsewhere. This stimulation causes a movement of the Receiver's hand, usually resulting in the "fire" key being hit on the Receiver's keyboard. This causes the cannon in the Sender's game to fire and destroy the pirate rocket before it hits the city.

Supply plane
Figure 6. In some trials, a friendly supply plane may move from right to left instead of a pirate rocket. The Sender must in this case rest rather than engage in imagery to make the cursor move away from the blue target and prevent any firing of the cannon at the supply plane.
Copyright 2013 R.P.N. Rao and A. Stocco