Spectral Feature Extraction and Analysis in Human Electroencephalogram (EEG) Signals
Mentor 1
Mohammad Rahman
Location
Union 340
Start Date
27-4-2018 1:20 PM
Description
Ongoing research is being conducted in the field of Brain-Computer Interfaces for use in technology for the rehabilitation of disabled individuals. These interfaces typically use electroencephalography as it allows individuals with spinal cord injury to control mechanisms despite a lack of muscle tone and control through the use of potential produced by brain activity. This study explores various methods for extracting spectral features from EEG data as electroencephalography has a very poor signal-to-noise ratio which necessitates the use of signal processing methods such as the Fast Fourier Transform and Discrete Wavelet Transform to extract usable and identifiable features, which are then used to train various machine learning algorithms to more quickly identify features and classify the data in the future. This study used various feature extraction methods based on the Fast Fourier Transform and Discrete Wavelet Transform to identify muscular flexion in a healthy adult male patient and assessed the most successful and viable methods. Some methods used were able to achieve accuracy percentages above 75%, with the most accurate (Fast Fourier Transform without summation) reaching 88.2% accuracy in differentiating muscle flexion from a relaxed state. This is a significant finding due to the difficulty of interpreting and classifying a non-linear and non-stationary signal such as those found in electroencephalography, as well as its applicability to the development of systems for assisting in the rehabilitation of the disabled. Additionally, this study explored the use of the Discrete Wavelet Transform in order to identify spectral features in muscle flexion which is a novel method for this specific application of identification that allows for the process of feature extraction and analysis to be completed using less processing power and time, allowing for more robust brain-computer interfaces. The methods explored in this study could allow for notable advances in prosthetics and rehabilitation technology.
Spectral Feature Extraction and Analysis in Human Electroencephalogram (EEG) Signals
Union 340
Ongoing research is being conducted in the field of Brain-Computer Interfaces for use in technology for the rehabilitation of disabled individuals. These interfaces typically use electroencephalography as it allows individuals with spinal cord injury to control mechanisms despite a lack of muscle tone and control through the use of potential produced by brain activity. This study explores various methods for extracting spectral features from EEG data as electroencephalography has a very poor signal-to-noise ratio which necessitates the use of signal processing methods such as the Fast Fourier Transform and Discrete Wavelet Transform to extract usable and identifiable features, which are then used to train various machine learning algorithms to more quickly identify features and classify the data in the future. This study used various feature extraction methods based on the Fast Fourier Transform and Discrete Wavelet Transform to identify muscular flexion in a healthy adult male patient and assessed the most successful and viable methods. Some methods used were able to achieve accuracy percentages above 75%, with the most accurate (Fast Fourier Transform without summation) reaching 88.2% accuracy in differentiating muscle flexion from a relaxed state. This is a significant finding due to the difficulty of interpreting and classifying a non-linear and non-stationary signal such as those found in electroencephalography, as well as its applicability to the development of systems for assisting in the rehabilitation of the disabled. Additionally, this study explored the use of the Discrete Wavelet Transform in order to identify spectral features in muscle flexion which is a novel method for this specific application of identification that allows for the process of feature extraction and analysis to be completed using less processing power and time, allowing for more robust brain-computer interfaces. The methods explored in this study could allow for notable advances in prosthetics and rehabilitation technology.
