Localization of Focal Intractable Epileptic Focus Based on Long-Term Recurrent Convolutional Network

In clinic, the success rate of surgical resection of the epileptic focus in the treatment of focal intractable epilepsy is only about 25%. The main reason for the low success rate is that the location of epileptic focal is difficult to accurately determine. Although some methods have built models to automatically identify the location of epileptic focal according to the characteristic differences between focal electroencephalogram (EEG) and non-focal EEG, the recognition accuracy of previous methods needs to be further improved. We note that the previous methods only consider the spatial differences and do not take the variation differences in the time domain into account. In this study, we intend to combine the convolution neural network and the long short-term memory neural network to construct a time-space deep learning model (long-term recurrent convolutional network, LRCN) to improve the accuracy of EEG recognition of epileptic focal. Based on the LRCN model, we achieved an accuracy of 97.13% for epileptic focal recognition, which performed out the existing methods.