EEG Signal Acquisition & IoT Event Detection

Goal

Design a Signal Acquisition System using the AD620 instrumentation amplifier, with downstream deep learning classification of EEG events and IoT-based remote analytics via AWS.

10-20 Electrode System

The 10-20 system is an internationally recognized method for placing electrodes on the scalp during EEG exams, sleep studies, and research. It ensures standardized and reproducible results by aligning electrode placement with specific brain regions.

• Standardization: Uses anatomical landmarks to ensure consistent electrode placement across subjects and studies
• Electrode Spacing: "10" and "20" refer to percentage distances between electrodes relative to total front-back or right-left skull measurements
• Anatomical Landmarks: Measurements taken from nasion to inion (front-back) and preauricular points (right-left)
• Electrode Labels: Letters represent brain regions (Fp = frontal pole, F = frontal); numbers indicate distance from midline (odd = left hemisphere, even = right)

Circuit Components

AD620

• Low noise: 9 nV/√Hz at 1 kHz — crucial for detecting small biosignals accurately
• High precision: 40 ppm maximum nonlinearity
• Low power: 1.3 mA max supply current — ideal for portable devices
• High CMRR: 100 dB min at gain = 10, rejecting common-mode interference
• Adjustable gain: 1 to 10,000 with a single external resistor
• Low input bias current: 1.0 nA max, suitable for high-impedance biosignal sources

AD822ARMZ

• Dual independent instrumentation amplifiers in a single package
• Adjustable gain from 1 to 10,000 using a single external resistor per channel
• High precision and low noise performance
• Suitable for simultaneous amplification and conditioning of multiple inputs

Data Cleaning: MATLAB Toolkit

• Artifact removal: EOG (eye movement and blinks), EMG (muscle activity), ECG (heart activity)
• Noise reduction: wavelet-based denoising
• Bad channel interpretation

Event/Task Identification using Mu Band

Mu rhythm is an EEG oscillation in the 8–13 Hz range, recorded from electrodes over sensorimotor cortex (typically C3, Cz, C4). Mu suppression is often interpreted as an indicator of mirror neuron system (MNS) activity.

Mu suppression occurs during:

• Execution of motor actions
• Observation of actions performed by others
• Motor imagery

Decreased mu power reflects increased cortical activation in sensorimotor areas. Mu suppression occurs before task.

Study Setup

• Rest/baseline EEG: monitored during rest or no activity
• Event: measured before, during, and after event for total of 10s
• Repeated for 10 different tasks across 8 subjects

EEG Data Analysis Pipeline

1. Clean and process data in 8s segments using MATLAB EEGLAB
2. Apply PyWavelets to get wavelet transform of time-variant EEG signals
3. Prepare data for deep learning — divide dataset into training and test sets; metadata includes baseline Mu power band
4. Run a 4-layer 1D CNN for binomial classification into rest and action using wavelet-transformed input
5. Analyze results — output is a trained model checkpoint and performance metrics

Code and dataset not publicly available due to the sensitive nature of the project.

IoT Application using AWS

• Remote data access: Real-time transmission of EEG data to healthcare professionals for remote monitoring and analysis
• Cloud-based storage: Secure storage of large volumes of EEG data, enabling long-term retention and multi-location access
• Advanced analytics: Cloud infrastructure enables machine learning algorithms on large datasets for pattern identification and outcome prediction
• Connected medical device: IoT-enabled EEG system integrates with other health monitoring systems for a comprehensive patient overview
• Enhanced research: Large, diverse EEG datasets accelerate neurological research and development of new treatments