Epileptic seizure prediction method through electroencephalogram signal on the basis of multi-scale convolution and self-attention network

Abstract

The invention discloses an epileptic seizure prediction method through an electroencephalogram signal on the basis of multi-scale convolution and a self-attention network. The epileptic seizure prediction method comprises the following steps that a convolutional neural network is combined with a long-short term memory network, an original electroencephalogram signal segment is taken as a network input, the idea of Inception is referred, a multi-scale convolution core is used to code an electroencephalogram signal sequence, a convolution operation is combined with pooling to complete down-sampling so as to keep features while dimensionality reduction is realized, lSTM is used for extracting the time sequence characteristics of the electroencephalogram signals, but the LSTM can only learn the front-to-back information of the electroencephalogram signals and but can not learn the back-to-front information, the bidirectional long-short-term memory network is combined with an attention mechanism to carry out modeling on the time characteristics of the electroencephalogram signal fragment, so that the influence of a complicated pretreatment process and manual intervention of scalp electroencephalogram signals is reduced, and better prediction performance is obtained. The epileptic seizure prediction method has certain generalization performance and can provide a certain basis for early warning of epileptic seizure.

Description

technical field [0001] The present invention belongs to the research field of predicting epileptic seizures. Specifically, multi-scale convolution is used to expand the sensory range of neurons to multi-channel raw signal data, which can obtain feature information of more scales, and use bidirectional attention network to model the temporal characteristics of EEG signal segments to reduce scalp EEG signal cumbersome preprocessing and the impact of human intervention for an end-to-end epileptic seizure prediction method. Background technique [0002] According to the report of the International League Against Epilepsy (ILAE), epilepsy is a chronic disease of temporary brain dysfunction caused by sudden abnormal discharge of brain neurons. one of the diseases. Seizures often produce disturbing physical symptoms such as uncontrolled twitching of limbs, loss of sensation or consciousness, disrupt patients' daily activities, and even increase the risk of premature death. For d...

AI Technical Summary

Problems solved by technology

At present, there are mainly the following methods in the research of predicting epileptic seizures: by calculating the time-frequency domain characteristics of EEG signal data, setting a threshold to distinguish preictal and interictal periods, but it requires a lot of manual feature extraction and does not have general Adaptability; and feature extraction and feature selection of EEG signal data, and then use traditional machine learning algorithms to complete the classification of EEG signals; some of the latest research uses deep learning technology to extract EEG signals by building network models feature and classify

However, how to achieve high sensitivity and low false prediction rate on the classification problem between interictal and preictal remains a major challenge

Images