This paper deals with seizure detection by processing EEG signals. In this context a methodology for transforming 1D EEG signals to 2D images thus the detection task can be accomplished by the Convolutional Neural Networks (CNNs) is proposed. The introduced method utilizing the high compactness of the Tchebichef moments along with the highly informative Recurrence Plots (RPs) that permits the application of pre-trained CNN models to detect the seizure cases. The proposed scheme provides improved detection accuracy up to 98% for the case of the Resnet18 model, while it shows outstanding robustness to additive (97%) as well as multiplicative (90%) noisy conditions. In this sense, the method outperforms the conventional approach of using the RPs to describe the raw EEG signals, by a factor of 1%-5%. These results are very promising and justify the efficiency of the introduced method, towards establishing a concrete and robust EEG signals analysis approach.
K. Tziridis, T. Kalampokas, G. A. Papakostas, “EEG Signal Analysis for Seizure Detection Using Recurrence Plots and Tchebichef Moments“, IEEE 11th Annual Computing and Communication Workshop and Conference (CCWC).