M.S. Thesis Defense: Xinyuan Ma
Friday, September 6, 2019
1146, A.V. Williams Building
301 405 3681
This work investigates the EEG signal processing and seizure detection based on deep learning architectures. The research includes two major parts. In the first part we use wavelet decomposition to process the signals as time frequency bands and extract signal features. The second part is the machine learning model and deep learning architecture we developed for seizure pattern analysis. The extracted feature maps are aligned as image inputs into our convolutional neural network (CNN) model. And we proposed our combined CNN-LSTM model to process the EEG signals as we have layers to function as feature extractors. In cross validation experiments, our CNN feature model could reach prediction accuracy of 96% and our CNN-LSTM model could reach 98% and 94% on average. In the developable work we proposed a matching network architecture which employs two parallel channels of our models to improve the sensitivity.