Seizure Prediction in Scalp EEG Using 3D Convolutional Neural Networks With an Image-Based Approach

Enhancing EEG-Based Classification of Depression Patients Using Spatial Information

“…Previous studies have shown that changes in entropy may reflect changes in brain activation when conducting cognitive tasks [41]. DE has been proven to be an effective feature in emotion recognition [36][37][38][39]. In this study, delta (1-4 Hz), theta (4-8 Hz), alpha (8-13 Hz), beta (13-30 Hz), and gamma (30-80 Hz) waves and wideband EEG (EEGW) were extracted by wavelet packet decomposition.…”

“…The earlier depression is detected, the easier it is to treat. As a low-cost, noninvasive acquisition, and high temporal resolution technique, electroencephalography is widely used in neural systems and rehabilitation engineering [5] [6]. Acharya et al proposed a typical computer-aided system for electroencephalogram (EEG)-based diagnosis of depression, which primarily includes an offline and online system [7].…”

“…[38][39][40][41][42]. Based on the filter bank common spatial pattern framework, Park et al proposed a filter bank regularized CSP…”

Seizure Prediction Using Multi-View Features and Improved Convolutional Gated Recurrent Network

“…Examples of these features include absolute and relative spectral band power [5], [6], Hjorth parameters [7], autoregressive coefficients [8], wavelet coefficients [9], largest Lyapunov exponent [10], statistical features [11], instantaneous amplitude and phase [12], phase correlation [13], phase synchronization [14], and common spatial pattern features [15]. Considering that seizure types and characteristics vary across patients and may evolve over time with the same patient, multi-view feature extractions are becoming necessary in EEG processing and have been employed in several studies [16], [17]. Yet efficient and reasonable combination of features is a crucial part and a challenging problem for seizure prediction.…”

“…At the same time, the time-varying information is essential for seizure prediction since the neural electrical activity is always a dynamic and nonstationary process. In recent studies, 3D CNN [16] and convolutional long short-term memory (C-LSTM) network [30] have been introduced to process sequences of feature matrices so that the fluctuation pattern of EEG can be recognized. To achieve a better prediction performance, we intend to propose an improved classification framework which is capable of analyzing sequences of multi-view feature matrices.…”