Date of Award
8-2017
Document Type
Thesis
Degree Name
Master of Science (MS)
Department
Electrical and Computer Engineering (Holcomb Dept. of)
Committee Member
Dr. Robert J Schalkoff, Committee Chair
Committee Member
Dr. Carl Baum
Committee Member
Dr. Brian C. Dean
Abstract
EEG is the most common test done by neurologists to study a patient’s brainwaves for pre-epileptic conditions. This thesis explains an end-to-end deep learning approach for detect-ing segments of EEG which display abnormal brain activity (Yellow-Boxes) and further classifying them to AEP (Abnormal Epileptiform Paroxysmals) and Non-AEP. It is treated as a binary and a multi-class problem. 1-D Convolution Neural Networks are used to carry out the identification and classification. Detection of Yellow-Boxes and subsequent analysis is a tedious process which can be fre-quently misinterpreted by neurologists without neurophysiology fellowship training. Hence, an au-tomated machine learning system to detect and classify will greatly enhance the quality of diagnosis. Two convolution neural network architectures are trained for the detection of Yellow-Boxes as well as their classification. The first step is detecting the Yellow-Boxes. This is done by training convolution neural networks on a training set containing both Yellow-Boxed and Non-Yellow Boxed segments treated as a 2 class problem, and is also treated as a class extension to the classification of the Yellow-Boxes problem. The second step is the classification of the Yellow-Boxes, where 2 different architectures are trained to classify the Yellow-Boxed data to 2 and 4 classes. The over-all system is validated with the entire 30s EEG segments of multiple patients, which the system classifies as Yellow-Boxes or Non-Yellow Boxes and subsequent classification to AEP or Non-AEP, and is compared with the annotated data by neurologists.
Recommended Citation
Ganta, Ashish, "Detection and Classification of Epileptiform Transients in EEG Signals Using Convolution Neural Network" (2017). All Theses. 2759.
https://open.clemson.edu/all_theses/2759