Research on Automatic Detection of Epileptic Seizures Based on the CNN-Transformer Model

Epilepsy is a severe brain disorder affecting approximately 50 million people worldwide. Given that the sudden onset of epileptic seizures can be quite serious, the detection of such seizures is crucial. Clinically, scalp electroencephalography (EEG) is commonly used as a non-invasive method to diagnose and identify the seizure foci in epilepsy, but manually interpreting EEGs is time-consuming and labor-intensive, which has driven the need for automated methods. To improve this, we have proposed a new framework that incorporates deep learning. This framework utilizes Convolutional Neural Networks (CNNs) to process multi-channel EEG data, extracting temporal and spatial information, while leveraging Transformer encoders to capture complex sequential data. On the public EEG epilepsy detection dataset TUSZ, our method achieved a weighted average F1 score of 0.778, as well as a precision of 0.808 and sensitivity of 0.769, showing significant results. Additionally, we have developed an epilepsy seizure detection system based on our model, which automatically assesses whether an epileptic seizure is occurring based on uploaded EEG files and implements warning responses.