Spike detection method based on temporal features and stacked bi-LSTM networks

Abstract

The invention discloses a spike wave detection method based on time series feature extraction and stacked Bi-LSTM network. Firstly, preprocessing is performed on the input original single-channel EEG signal, and the preprocessed EEG signal is segmented to obtain EEG signal fragments. Smooth nonlinear energy features and morphological features are obtained through two time series feature extraction algorithms. After cutting the obtained two time series features to ensure that the length is consistent with the EEG signal segment, the feature matrix is ​​spliced ​​with the EEG signal segment, and then the stacked Bi‑LSTM network model is trained using the obtained feature matrix and labeling information; The test data is used to test the trained stacked Bi-LSTM network model, and optimize the model performance according to the test results. The method of the present invention effectively learns the temporal sequence characteristics of the EEG through the cyclic neural network model to achieve the effect of accurately detecting the spike discharge; and can simultaneously detect the spike and its generation channel position.

Description

technical field [0001] The invention belongs to the field of electroencephalogram signal processing, and relates to a spike wave detection method based on time series feature extraction and stacked bidirectional long-short-term memory (stack Bi-LSTM) network. Background technique [0002] Epilepsy is a common chronic neurological disease that seriously threatens the life and health of children and adults. Spikes and their complex waveforms are the pathological basis of epileptic seizures. The relevant parameters such as spike discharge time and discharge position are of great significance. The first step to determine these parameters is spike detection. [0003] The existing spike detection methods are mainly divided into two types. The traditional method is often biased towards feature engineering. After designing one or more features that can characterize the characteristics of spikes, the signals are classified into spikes by threshold method and other relatively simple c...

AI Technical Summary

Problems solved by technology

[0004] 1. Due to the weak performance of the classifier in the traditional method, the characterization ability of the extracted features is relatively harsh. In order to improve the performance of the model, it is often necessary to extract multiple features, and the computational complexity is high;

[0005] 2. Deep learning methods generally use fully connected neural networks or convolutional neural networks, which cannot effectively use the characteristics of EEG signals as a sequential signal

Images