DSCNN-LSTMs: A Lightweight and Efficient Model for Epilepsy Recognition

• DOI: 10.3390/brainsci12121672
• Publication Date: 2022-12-05T12:07:31Z
• Authors: Zhentao Huang; Yahong Ma; Rongrong Wang; Baoxi Yuan; Rui Jiang; Qin Yang; Weisu Li; Jingbo Sun
• Keywords: Binary classification; Feature (linguistics)

Epilepsy is the second most common disease of nervous system. Because its high disability rate and long course disease, it a worldwide medical problem social public health problem. Therefore, timely detection treatment epilepsy are very important. Currently, professionals use their own diagnostic experience to identify seizures by visual inspection electroencephalogram (EEG). Not only does require lot time effort, but process also cumbersome. Machine learning-based methods have recently been proposed for detection, which can help clinicians make rapid correct diagnoses. However, these often extracting features EEG signals before using data. In addition, selection requires domain knowledge, feature types significant impact on performance classifier. this paper, one-dimensional depthwise separable convolutional neural network short-term memory networks (1D DSCNN-LSTMs) model epileptic autonomously raw EEG. On UCI dataset, 1D DSCNN-LSTMs verified cross-validation complexity comparison. Compared with other previous models, experimental results show that highest recognition rates binary quintuple classification 99.57% 81.30%, respectively. It be concluded in paper an effective method based signals.