Early Prediction of Epileptic Seizure Based on the BNLSTM-CASA Model

Abstract

Epilepsy is one of the world's most common neurological diseases. Reliable early prediction and warning of seizures can provide timely treatment for patients with epilepsy, and improve their quality of life. Compared with most hand-designed prediction methods, an automatic prediction model that can process the original electroencephalogram (EEG) signals directly and take into account the leads optimization problem is needed. In this paper, we proposed an end-to-end automatic seizure prediction model based on the Batch Normalization Long Short Term Memory networks (BNLSTM) and Channel and Spatial attention (CASA). Firstly, raw EEG signals without any preprocessing are used as the input to the system, which can reduce the computation amount. Secondly, BNLSTM and CASA retained the time and spatial information of the raw EEG data respectively. Channel attention (CA) achieved the automatic optimization of EEG full-lead data and improved the prediction accuracy. Spatial attention (SA) achieved the adaptive learning of feature parameters. Finally, a fully connected layer is applied to predict the seizures. The performance of the seizure prediction model we proposed is evaluated on the data of 14 patients with Area Under the Curve (AUC) of 0.986, accuracy(Acc) of 0.956, specificity(Spe) of 0.968, and sensitivity(Sen) of 0.942. In addition, the proposed method provided an accurate prediction for all 50 seizures of the other 5 patients in the generalization dataset. Experimental results show that the proposed model has a certain generalization performance, which can provide a reliable basis for early warning of epileptic seizures.

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

“…Additionally, the procedure may be required in some applications such as wearable devices where using a large number of channels is impractical [ 26 ]. Channel selection can be performed using different approaches, whether they are statistical approaches [ 11 , 22 , 36 , 62 , 75 , 76 , 113 ], data-driven approaches [ 14 , 88 , 116 , 117 , 118 ], wrapper approaches [ 119 ], or from prior knowledge based on previous studies [ 120 , 121 ].…”

“…Other data-driven approaches involve using attention mechanisms that can automatically attend to informative channels by assigning attention scores to these channels during the model’s training process. In [ 117 ], maximum and average pooling operators are applied to the input feature matrix, followed by a convolution layer. The outputs are vertically concatenated, and a non-linear sigmoid activation function is then applied to form the channel attention score.…”

“…Throughout the survey, various strategies were implemented to improve the ability of models to generalize across data. This includes, but is not limited to, different regularization techniques [ 117 ], early stopping [ 58 , 123 ], data augmentation [ 50 ], ensemble learning [ 90 ], transfer learning [ 22 ], multi-view learning [ 19 , 57 , 124 ] and subject-independent approaches [ 44 , 101 , 125 , 126 ]. Since discussing the approaches to regularization is a broad topic, it is considered beyond the scope of this review.…”

“…In [ 117 ], the batch normalized LSTM (BN-LSTM) technique was used. The difference between LSTM and BN-LSTM is that the latter handles covariance shifts that happen between the hidden-to-hidden layers, boosting the model’s generalization and convergence [ 129 ].…”

A Spatiotemporal Graph Attention Network Based on Synchronization for Epileptic Seizure Prediction

“…Noting that current studies of EEG signals have focused on time-frequency analyse and ignored the influence of spatial factors, Ma et al [20] introduced the channel and spatial attention (CASA) into batch normalization long short-term memory (BNLSTM) [21] to preserve the spatial and temporal information of EEGs. Sun et al [22] proposed the channel attention dual-input convolutional neural network, which can synthesize the time-domain, frequency-domain and spatial information of EEG data to achieve accurate seizure prediction of real, representable EEG signals.…”