Electronic Theses and Dissertations


Saleha Khatun



Document Type


Degree Name

Doctor of Philosophy



Committee Chair

Bashir Morshed

Committee Member

Gavin Bidelman

Committee Member

Eddie Jacobs

Committee Member

Aaron Robinson


Electroencephalogram (EEG) is a technique for recording asynchronous activation of neuronal firing inside the brain with non-invasive scalp electrodes. EEG signal is well studied to evaluate the cognitive state, detect brain diseases such as epilepsy, dementia, coma, autism spectral disorder (ASD), etc. In this dissertation, the EEG signal is studied for the early detection of the Mild Cognitive Impairment (MCI). MCI is the preliminary stage of Dementia that may ultimately lead to Alzheimers disease (AD) in the elderly people. Our goal is to develop a minimalistic MCI detection system that could be integrated to the wearable sensors. This contribution has three major aspects: 1) cleaning the EEG signal, 2) detecting MCI, and 3) predicting the severity of the MCI using the data obtained from a single-channel EEG electrode. Artifacts such as eye blink activities can corrupt the EEG signals. We investigate unsupervised and effective removal of ocular artifact (OA) from single-channel streaming raw EEG data. Wavelet transform (WT) decomposition technique was systematically evaluated for effectiveness of OA removal for a single-channel EEG system. Discrete Wavelet Transform (DWT) and Stationary Wavelet Transform (SWT), is studied with four WT basis functions: haar, coif3, sym3, and bior4.4. The performance of the artifact removal algorithm was evaluated by the correlation coefficients (CC), mutual information (MI), signal to artifact ratio (SAR), normalized mean square error (NMSE), and time-frequency analysis. It is demonstrated that WT can be an effective tool for unsupervised OA removal from single channel EEG data for real-time applications.For the MCI detection from the clean EEG data, we collected the scalp EEG data, while the subjects were stimulated with five auditory speech signals. We extracted 590 features from the Event-Related Potential (ERP) of the collected EEG signals, which included time and spectral domain characteristics of the response. The top 25 features, ranked by the random forest method, were used for classification models to identify subjects with MCI. Robustness of our model was tested using leave-one-out cross-validation while training the classifiers. Best results (leave-one-out cross-validation accuracy 87.9%, sensitivity 84.8%, specificity 95%, and F score 85%) were obtained using support vector machine (SVM) method with Radial Basis Kernel (RBF) (sigma = 10, cost = 102). Similar performances were also observed with logistic regression (LR), further validating the results. Our results suggest that single-channel EEG could provide a robust biomarker for early detection of MCI. We also developed a single channel Electro-encephalography (EEG) based MCI severity monitoring algorithm by generating the Montreal Cognitive Assessment (MoCA) scores from the features extracted from EEG. We performed multi-trial and single-trail analysis for the algorithm development of the MCI severity monitoring. We studied Multivariate Regression (MR), Ensemble Regression (ER), Support Vector Regression (SVR), and Ridge Regression (RR) for multi-trial and deep neural regression for the single-trial analysis. In the case of multi-trial, the best result was obtained from the ER. In our single-trial analysis, we constructed the time-frequency image from each trial and feed it to the convolutional deep neural network (CNN). Performance of the regression models was evaluated by the RMSE and the residual analysis. We obtained the best accuracy with the deep neural regression method.


Data is provided by the student.

Library Comment

Dissertation or thesis originally submitted to ProQuest