Brain-Computer Interface: Review

Brain Computer Interface: A Review

• A brain-computer interface (BCI) facilitates communication between the human brain and a computer system through non-muscular channels.
• Advancements in electronics and the need to assist individuals with neuromuscular disorders have spurred the development of BCI.
• Electroencephalogram (EEG), which records electrical activity from the scalp, is used to characterize brain activity.
• EEG measures brain activity non-invasively using portable equipment.
• BCI systems translate EEG activity into commands for controlling external devices like robotic arms and wheelchairs.
• Standard BCI setups involve data acquisition, signal pre-processing, feature extraction, classification, and control interface.

Introduction

• Cognitive neuroscience and brain imaging advancements enable communication between brain activity and peripheral devices.
• BCI technology monitors and controls activity in real-time, benefiting individuals with disabilities.
• BCI assists with biofeedback training for conditions such as stroke, epilepsy, and ADHD.
• Sensors record changes in electrical potential, magnetic fields, and ion flow related to brain activity.

What is BCI?

• Brain-Computer Interface (BCI) provides a means of communication for people with movement disabilities using EEG or other brain signals.
• The brain contains over 100 billion neurons, which are studied to understand the signals they produce.
• Researchers are focusing on applications such as light/television control, text processing, and virtual reality games.

Early History

• Richard Caton first recorded EEG signals from animal cortical surfaces in 1875.
• Hans Berger discovered electrical signals from the human brain using EEG in 1929 and recorded Alpha Waves (8-10 Hz).
• Adrian and Matthew discovered that EEG signals vary across the head and suggested standardized electrode positioning.
• The USA's Defence Advanced Research Projects Agency (DARPA) started a program in the 1970s to explore EEG activity, expanding research beyond clinical diagnosis.

Advancement

• Research on BCI increased significantly after 1995 due to technological advancements.
• BCI Information Transfer Rates (ITR) improved from 5-25 bits/min to 84.7 bits/min.
• BCI advancements aim to improve the lives of 'locked-in' patients by providing means of communication.
• Research is geared towards a greater understanding of EEG and enhanced brain activity recording techniques.
• Advancements in low-cost microelectronics enable complicated tasks through embedded circuits.
• Machine learning algorithms contribute to brain-controlled applications.

Brain Computer Interface: Basic

• Neurons, the micro-processing stations in the brain, have input, trigger, conduction, and output functionalities.
• BCI uses electrical signals generated by neuron firing or inhibition.
• Neurons fire or inhibit based on whether data severity exceeds or falls below a threshold.
• The brain consists of 52 discrete points in a cytoarchitectural map.
• A neuron activates on the scalp as an electrical pulse or magnetic field during various activities.
• A BCI system converts input from the brain into real-time action.
• Key BCI phases include data acquisition, signal processing & classification, computer interface, and application.

Data Acquisition

• In BCI, signal acquisition is done using invasive, partially invasive, and non-invasive methods.
• The invasive technique involves reading brain signals from inside the brain's gray matter requiring brain surgery.
• Electrocorticography (ECOG) is a partially invasive BCI method, done by implanting a BCI device outside the grey matter.
• Noninvasive BCI is the most used method in which electrodes are mounted on the scalp.
• EEG signals are noisy compared to invasive methods.
• Non-invasive methods include Magnetoencephalography (MEG), Magnetic Resonance Imaging (MRI), Functional Magnetic Resonance Imaging (fMRI) and P-300 based BCI, and Positron Emission Tomography (PET).
• fMRI measures hemodynamic response to neural activity and is popular due to its low invasiveness, absence of radiation exposure, and wide availability.
• MEG uses superconducting quantum interface devices (SQUIDs) arrays to record magnetic fields.
• EEG is used for real-time applications because it has a short time constant and easy-to-use equipment.
• The American EEG society uses the International 10-20 system to standardize the position of the electrode for use.
• EEG signals can be acquired using wet Ag/AgCl electrodes.
• Conventional wet electrodes require preparation time, conductive gel, and may cause allergic reactions.
• Microelectromechanical Systems (MEMS) helps to design dry MEMS electrodes to acquire EEG signals.
• Recording brain activity requires mounting multiple electrodes on the scalp for 20-40 minutes per run.
• Active electrodes collect and amplify signals, and voltage drop occurs in passive electrodes.

Signal Processing and Classification

• Non-invasive EEG acquisition captures input signals, contains noise, electronics interference, electromyography (EMG) and ocular artifacts, which can lead to incorrect results.
• Signal processing enhances acquired input signals and involves time domain or frequency domain techniques.
• Bandstop filters attenuate frequencies in the range of 50 to 60 Hz, though there are some limitations.
• Useful information within the EEG lies within Theta band 4-8 Hz, Alpha-1 band 8-10 Hz, Alpha-2 band 10-20 Hz, Beta band 12-30 Hz, and Gamma band 30-100 Hz
• Bandpass filters extract specific frequency ranges.
• Noise filters adapt to the spectrum of an input signal and attenuates an input signal in frequency.
• Adaptive filtering helps the EOG or EMG has a strong correlation with measurement.
• Blind Source Separation identifies EEG sources and incorporates EOG/EMG signals into analysis.
• Independent Component Analysis (ICA) decomposes noise, assuming sources equal electrodes and no time delay.
• Centering and Whitening are major steps of ICA.
• Matching Pursuit Algorithm decomposes signals into Gabor dictionary components and can compared to algorithms by Barreto and etc.
• Recurrent quantum neural network(RQNN) can be used to filter.

Computer Interface

• Recorded EEG signals can be processed online or offline for feature extraction along with computer algorithms.
• Feature extraction from EEG signals includes amplitude or latencies of potentials, or frequencies of rhythms.
• Systems extract features and convert them into meaningful control using translation algorithms, which can be simple or complex neural networks.
• Users adapt algorithms, as well.
• EEG signals are affected by hormonal levels, environment so BCI requires adaptation.

BCI Characteristic

• BCIs are categorized as dependent or independent.
• Dependent BCIs use activity to drive the process.
• Independent BCIs rely on brain's normal output pathways.
• Independent BCIs are used for those with severe neuromuscular disability.

Skill in BCI

• BCI evolves to reflect the brain's thought process via EEG signals and helps to understand the mind.
• Electrophysiology signals change EEG rhythms and neural firing for control.
• The brain neuromuscular output channels depend on feedback.
• The brains adaptive capacity can be extended to control electrophysiological signal.

The mental strategy

• Controlling computer peripherals involves self-controlling brain cortical potential so seek the desired brain potential.
• Motor imagery triggers cortical areas similar to executing the same movement.
• Mental tasks such as concentrating on lights, flashed letters, mental arithmetic, and imaging rotational objects are appropriate for modulating signals.

Brain-Computer Interface: Challenges

• BCI development depends on signals, data acquisition, feature extraction, translation algorithms, output devices, modes, training, protocols and use groups.
• Long-term stability is questioned, if neuronal activity can function without movement.
• The users ability to use BCI varies.
• Translation algorithms convert user input into control output, depending on signal features.
• Further study is required in BCI aspects including signal preprocessing, feature extraction, translating algorithm, and user interfaces.

Tools in BCI Research

• Numerical data analysis tools such as Matlab, Octave, and Scilab are available for BCI research, as well as BioSig.
• Software provides similarities and dissimilarities in data processing methods, but BioSig helps make research efficient.

Application

• Robotics prosthetic or an exoskeleton with brain control are the applications related to mobility.
• First BCI to restore full body mobility, Walk Again Project in development.
• With direct mouse, monitor and keyboard communication BCI has potential to insert a user into virtual world.
• Emotive device is example of BCI.
• Environment control, locomotion and exercise are five potential BCI application.
• EEG analysis and machine learning techniques is used in mobile robot control.
• BCI has applications such as the assisting communication, interaction with the patients suffering from neurological disease.
• A parameter required to extract specific frequency band in autoregressive spectral analysis.

Conclusion

• This survey focused on BCI components, signal acquisition, feature identification, algorithm with tools such as GNU Octave, FreeMat, Scilab and MARLAB.
• BCI is used for physically disabled.
• BCI varies as per application and algorithms.
• EEG signals can be recorded by invasive and noninvasive methods.
• Microelectromechanical system and Nanotechnology-based invasive methods is open BCI research.
• In addition, research is done to improve smooth control.
• There is a lot of scope in the field of BCI.
• Research is in the inter-disciplinary field for neuroscientists, engineers, psychologists, and computer scientists.