Project

EEG sensor data analysis for mental health assessment

A very interesting field of Computer Science called Artificial Intelligence is concerned with the study and design of intelligent machines and applications. Advanced AI technologies are already at work around us in almost all areas, and medical science is not left behind. The relationship between Computer Science and Medical Science is not new, there are thousands of computer aided diagnostic tools available across the world to identify patient’s health issues. A common and burdensome problem concerned with human beings is mental health or psychological disorders such as depression. The symptoms of mental health, specifically depression, appear mostly as behavioral. Studies say that Electroencephalogram (EEG) may be used as a tool to diagnosis a human’s mental or behavioral health. The behavioral and mental healthcare fields are also benefitting from advancements in AI. 
 
 The objective of this study is to assess the human’s mental health based on 5 main brainwave signals collected using an EEG sensor. The objective of this study was twofold, first to collect the brain signals data of a human being in their different states of mind, and second using those data to train Machine Learning and Deep Learning techniques. The complex, non-linear and non-stationary EEG signals are very tedious to interpret visually and it is highly difficult to extract the significant features from them. Hence, nonlinear dynamic methods are used in extracting the EEG signal for computer aided diagnosis of mental health. Data has been collected for some time based on the specified mood status which are considered as dependent on each other within a certain time period. A powerful type of neural network called Recurrent Neural Network which is designed to handle sequence dependence, is used in this study, as well as the Long Short-Term Memory network or LSTM network. Prediction had the highest accuracy using standard neural network when focusing on only happy/ sad states, but accuracy varied with LSTM using ‘tanh’ and ‘softmax’ activation functions.

Project (M.S., Computer Science)--California State University, Sacramento, 2017.

A very interesting field of Computer Science called Artificial Intelligence is concerned with the study and design of intelligent machines and applications. Advanced AI technologies are already at work around us in almost all areas, and medical science is not left behind. The relationship between Computer Science and Medical Science is not new, there are thousands of computer aided diagnostic tools available across the world to identify patient’s health issues. A common and burdensome problem concerned with human beings is mental health or psychological disorders such as depression. The symptoms of mental health, specifically depression, appear mostly as behavioral. Studies say that Electroencephalogram (EEG) may be used as a tool to diagnosis a human’s mental or behavioral health. The behavioral and mental healthcare fields are also benefitting from advancements in AI. The objective of this study is to assess the human’s mental health based on 5 main brainwave signals collected using an EEG sensor. The objective of this study was twofold, first to collect the brain signals data of a human being in their different states of mind, and second using those data to train Machine Learning and Deep Learning techniques. The complex, non-linear and non-stationary EEG signals are very tedious to interpret visually and it is highly difficult to extract the significant features from them. Hence, nonlinear dynamic methods are used in extracting the EEG signal for computer aided diagnosis of mental health. Data has been collected for some time based on the specified mood status which are considered as dependent on each other within a certain time period. A powerful type of neural network called Recurrent Neural Network which is designed to handle sequence dependence, is used in this study, as well as the Long Short-Term Memory network or LSTM network. Prediction had the highest accuracy using standard neural network when focusing on only happy/ sad states, but accuracy varied with LSTM using ‘tanh’ and ‘softmax’ activation functions.

Relationships

Items