An Attention-Based Wavelet Convolution Neural Network for Epilepsy EEG Classification

Abstract

As a kind of non-invasive, low-cost, and readily available brain examination, EEG has attached significance to the means of clinical diagnosis of epilepsy. However, the reading of long-term EEG records has brought a heavy burden to neurologists and experts. Therefore, automatic EEG classification for epileptic patients plays an essential role in epilepsy diagnosis and treatment. This paper proposes an Attention Mechanism-based Wavelet Convolution Neural Network for epilepsy EEG classification. Attention Mechanism-based Wavelet Convolution Neural Network firstly uses multi-scale wavelet analysis to decompose the input EEGs to obtain their components in different frequency bands. Then, these decomposed multi-scale EEGs are input into the Convolution Neural Network with an attention mechanism for further feature extraction and classification. The proposed algorithm achieves 98.89% triple classification accuracy on the Bonn EEG database and 99.70% binary classification accuracy on the Bern-Barcelona EEG database. Our experiments prove that the proposed algorithm achieves a state-of-the-art classification effect on epilepsy EEG.

Supervised Machine Learning and Deep Learning Techniques for Epileptic Seizure Recognition Using EEG Signals—A Systematic Literature Review

“…DWT is one of the most widely applied wavelet transforms in the field of EEG signal analysis. In [ 20 , 35 , 63 ], DWT was used to obtain most of the brain’s rhythmic frequencies. Wavelet coefficients, named detail and approximation coefficients were extracted to capture the low and high frequencies in the wavelet with a varying number of filter banks.…”

“…This procedure results in weighted feature maps corresponding to different channels, where features with high weights are considered more important to the seizure prediction task. A similar concept was introduced in [ 63 , 116 ], but a non-linear softmax activation function was used instead of a sigmoid one. Likewise, in [ 118 , 122 ], the mechanism attends to different channels by learning the attention weights through incorporating multiple layers of convolution layers followed by fully connected layers.…”

EEG-Based Seizure Prediction via Model Uncertainty Learning

“…Different from numerous methods [5], [9], [16], [18] . The SPH [33] refers to the intervention period before seizures, where therapies (such as electrical stimulation) could be performed. The SOP indicates the period when seizures are anticipated to occur, which equals the preictal period in duration.…”

“…Many literatures [5], [9]- [14] have shown that leading performance could be achieved with DNNs for seizure prediction, in contrast to traditional machine learning methods. Several studies [9], [12], [15], [16] have proposed methods to manually extract features from complex raw EEG, which are widely used to eliminate EEG artifacts. Khan et al [15] processed the raw EEG signal using wavelet transform.…”

Multilevel Feature Learning Method for Accurate Interictal Epileptiform Spike Detection

“…Monitoring brain neuron electronic activity is crucial for the treatment of neurological disorders. A routine scalp electroencephalogram (EEG) is considered a better modality for analyzing brain activities [3], [4]. Studies have shown that EEG signals in the interictal period can be measured as the epileptogenic biomarker-interictal epileptiform spike (herein referred to as spike) [5], [6].…”