SKINTRONICS: Brain-machine interfaces (BMI) enabled by flexible scalp electronics and deep-learning algorithm

Overview

Problem / Solution

Modern electroencephalography (EEG) systems offer non-invasive, real-time monitoring of brain electrical activity with good resolution and low cost. EEG-based brain-machine interfaces (BMI) are used for rehabilitation by controlling prosthetic systems and improving the quality of life for people with motor disabilities.

 However, current EEG systems are often burdened with the following issues: bulky and heavy, require numerous electrodes, extensive preparation time, require conductive gels, regular maintenance, impedance variations due to scalp hair, and motion artifacts. Present EEG classification methods require per-subject or per-session training due to the variation in human brains. Hence the difficulties in implementing EGG into BMI. 

 SKINTRONICS is a fully portable, wireless, and flexible scalp EEG monitoring system that uses Steady-State Visually Evoked Potentials (SSVEP) to control machines via Bluetooth. It features a simple and compact design, maximum comfort, a short set-up time, and improved performance.

Design

SKINTRONICS is a fully portable, wireless device that accurately monitors steady-state visually evoked potentials (SSVEP) on the scalp in real-time for a brain-machine interface (BMI). SKINTRONICS works by detecting the EEG signals from two channels via the electrodes. Then, these signals are amplified and run through an analog-to-digital converter (ADC). The microcontroller unit (MCU) transmits the data via the Bluetooth telemetry unit. As a computer receives the data, it is preprocessed with filtering, processed with the trained deep neural network algorithm, and then transferred to the control target.

Steady-State Visually Evoked Potentials (SSVEP)

Capturing and classifying SSVEP is a potential therapeutic BMI strategy. The subject is asked to focus on flickering stimuli for this type of interface, while an EEG system records brain electrical activity from specific locations on the scalp. Because these stimuli elicit frequency-dependent brain activity, arrays of stimuli may be used as an interface for subjects to gaze between in order to control some target.

Optimal Electrode Locations 

The three optimal electrode locations (O1, O2, and Oz) are determined through a deep convolutional neural network analysis of the 32-channel EEG recording of SSVEP data. The data is sampled at 256 Hz and the reference was set at Oz. The electrode positions of O1, Oz, and O2 have the largest signal weights and a consistently high signal-to-noise ratio (SNR). These electrodes are the most useful for gathering EEG signals. Two channels (O1–Oz and O2–Oz) offer minimal contact area for a comfortable, dry EEG recording.

Hardware 

SKINTRONICS is comprised of three primary components: a flexible wireless membrane circuit on the back of the neck, a printed skin electrode on the mastoid, and three flexible hairy scalp electrodes on the occipital lobe. These components offer reliable mechanical flexibility and stretchability. The EEG recording setup for two channels incorporates the skin-like electrode and elastomeric electrodes. 

Nanomembrane Electrode

The ultrathin membrane electrode placed on the right mastoid behind the ear serves as driven ground. This aerosol-jet printed, skin-like electrode features an open-mesh structure for stretchability and flexibility.

Elastometric Scalp Electrodes 

Three elastomeric electrodes secured on a headband are in direct contact with the scalp above the occipital lobe. These dry electrodes have eight flexible elastomer legs that gently splay when downward pressure from the headband is applied. The splaying action of the Ag/AgCl tipped legs moves and separates the hairs, thus allowing for effective scalp contact. High skin impedance of less than 20 kΩ is achieved due to conformal skin contact without the use of gel or adhesive that reduces the noise and interference in signal recording.

The dry electrodes and the circuitry are linked using PDMS-insulated conductive film cables and attached using silver paint as an adhesive.

Ultrathin Wireless Electronics

The electronic system placed behind the neck is composed of different surface mount chip components, including a Bluetooth telemetry unit. The thin-film miniaturized flexible circuit boards are optimally embedded on a polydimethylsiloxane (PDMS)-coated silicon wafer constructed using a microfabrication process. The assembled circuit is completely encapsulated with a soft low-modulus elastomeric membrane to protect the circuit components and to provide the required adhesiveness and compliance for skin application. 

Deep Learning

Detecting and analyzing SSVEP signals are difficult due to low signal amplitude and also variations in human brains. The SSVEP signal amplitude within the range of tens of micro-volts is similar to electrical noise in the body. A deep learning neural network algorithm filters and analyzes these signals to determine exactly the user’s intention.